While I don’t plan to speculate in this essay about the longer horizon of full-automation (I’m working on a larger essay Post-”Labour”/”AGI” economics) - I do want to talk about the more immediate 3-10 years.

A lot of people online expect AI tools to elevate everyone. However, from personal experience working with and talking to many people who are using AI today, I have come to a completely different conclusion.

A person needs to hit a threshold of specific personality traits, like agency, intelligence, agility, and context-switching, just to have net-positive returns from using AI!

Then, if you are past this threshold, the difference between 2x, 3x, 5x returns and 100x returns from using these tools is an insanely different bar. AI multiplies talent exponentially.

In far too many cases, mid-talent might have net-NEGATIVE returns for an organization, as they cost you attention, time, and energy from your other top talent, all WHILE not providing value fast enough. The tasks they are able to work on are easily done by others with AI as an afterthought, and having them do those tasks actually just slows the company down, all while they burn shitloads of tokens and increase the org’s costs.

In the old economy, companies needed large layers of average workers because work had to be decomposed, coordinated, reviewed, and executed by humans. In the AI economy, more of that middle layer gets absorbed by a small number of high-agency operators using tools. So the value of “just doing the task” collapses, while the value of judgment, taste, agency, distribution, trust, and ownership rises.

These dynamics, in my opinion, will result in two extremes in the future economy:

1) Top 0.1% talent in any field will become extremely valuable.

We are already seeing this play out in the Silicon Valley talent war, with millions, and even tens of millions of dollars, in compensation packages for engineers.
That signals one of two things: (1) we are in a crazy bubble that is about to pop, or (2) the top 0.1% of a specific field might provide 99% of the value created, and “overpaying” for this talent compared to the average labour market rate for those roles is worth it.

While I believe the first is plausible, I am convinced we are in the latter scenario.

And while the top 0.1% of any field and in any company providing 99% of the value created, then the question arises - do you need the rest? especially if they are not even neutral-returns, but negative ones?

Probably not. Which bring us to the other extreme of the economy:

2) As AI concentrates economic production among top talent / high-agency operators, I think we’ll see others respond defensively by forming medieval guild-like structures to preserve income, social dignity, local relevance, and bargaining power.

The first is work that depends on a physical area: things tied to physical space, homes, neighborhoods, land, maintenance, lawn mowing, local services, and so on. These are businesses where proximity still matters, where being part of the local community still matters, and where trust is not easily replaced by software.

The second is work that is deeply community driven - businesses and services that flourish through word of mouth, local reputation, shared identity, or one trusted member of the community. In a world where digital production becomes increasingly automated and abundant, local trust becomes a kind of economic moat.

The third is human-experience work: IRL events, art, street art, theater, workshops, classes, performances, shared meals, physical spaces, and things that feel “humans by humans.” Anything connected to the physical world, anything that is not purely digital and can be human-improved, will become more valuable.

People will seek more local experiences with people they know, and with communities they feel part of. They will want things that are not just optimized, generated, and infinitely scalable. They will want things that feel human, situated, embodied, and socially real.

—

If this thesis is even directionally right, the personal implications are uncomfortable.

Therefore, my generic take is that you should work as hard as you can to build assets. Own businesses, property, distribution, audience, relationships, land, invest in Google, or anything else that can compound outside of wages. Wages, as we know them today, are not the way to enrich yourself in these turbulent times, and they may not be a reliable source of stability in the future.

It may also become harder to get a job than to invent one. We are already seeing college students say it feels easier to build and sell your own product than to find an entry-level job. I think this becomes more common. So start building your own stuff, sharpen your agency. Or - find a way to provide value to people in your community more directly directly, can be on smaller scales. In the long run, this may become a much better way to make ends meet.

The uncomfortable part is that I don’t think there will be a clean place for everyone in the old middle of the labour market. Some people will become hyperscalers, using AI to produce at insane leverage. Others will move toward guilds, local economies, physical work, trusted communities, human experiences, and things that still require humans in the loop.

The positive take here, is that I don’t think this means humans have no value. I think it means human value will shift away from just “doing the task,” and toward ownership, agency, taste, trust, relationships, physical presence, and building things that other humans actually want to be part of.

Adam

Well - the first and most obvious reason - is that I love spending time with fellow Jews, who share the same traditions, the same connection to our beloved Israel, and who can understand me slightly better than others.

Secondly, after experiencing a massive antisemitic backlash against me last September for launching the AI Key, as well as observing the growing hatred appearing plainly on the streets around the world (in SF, Prague, London - everywhere I go - see pictures at the end). I felt the need to build a stronger community of people around me who truly understand this challenge.

Thirdly, I thought to myself - what would be the worst outcome for them, for their antisemitism against me? Well - if it made us stronger, more united, and more helpful to one another than ever before - that will do for now. Let them keep pushing, so that we can become an even shinier diamond. But we mustn’t let them stop us from creating value in this world.

Fourthly, I miss home. I miss Israel, my family, the people, the energy. Yet, I can’t go back right now. My purpose here has yet to be completed.

And lastly, maybe the most important of all: right now, San Francisco and the Bay Area is the most critical pressure point of one of the greatest revolutions of our lifetime, if not the greatest for generations to come. Greater than the Industrial Revolution, the printing press, the internet, mobile, everything. The founders and the leaders of the future are all passing through this little piece of land on the West Coast of the United States. Everything that is our species’ future is being developed here. From AI, robotics, space exploration, automation, new economics and governance structures, human augmentation and BCI, philosophy and the human condition, medicine, longevity, simulations, arts and culture, energy, biotechnology, to commerce.

We traditionally have contributed, and must continue to contribute, to every single one of those disciplines. We cannot be isolated in our little part of the world - we must exist, meet, and participate where the action is happening. Those great people who will shape humanity for decades to come must be familiar with us, we must contribute and work with them - and we must learn from the frontier so we can be the frontier. We cannot allow the hatred against us, and the push against our economy, to sabotage our participation in this great revolution. We must stand strong, create value, and collaborate with everything that is happening here.

IsraValley is my little attempt to build a beneficial and strong bridge between Silicon Valley and Israel. To allow a flow of Israeli entrepreneurs to come to the city and experience it to its fullest. To learn and bring it home. To come and build here as well, so we can have a great share of the future, and to befriend the leaders of these changing times. As well as to allow the founders and companies of the Valley to be accessible to the great talent of Israel, the amazing people, and to incentivize working relationships with our companies.

So on my 27th birthday, I write it here, to say it plainly - a great future awaits, and we will all build it together, layer by layer, in the Promised Land and in the Valley.

- Adam Cohen Hillel
IsraValley.com

The following pictures are only from 100 meter walk from my house to the nearest Cafe. These includes sentences like “by all means necessary”.

I think there’s truth to it - but rather than being the fundamental force - it is one face of it, at least when it comes to humans. It is like describing the brain with only the amygdala, the preemptive system of our brain, while ignoring the cortex, and higher level, more modern part of it.

So, on top (or rather, next to) the will to power - lies another fundamental force that shapes our society. A society that its existence is rooted in the unknownness of the universe.

We were gifted with consciousness to observe and reason about our own existenceand about the universe - yet, from our limited observation - we are the only ones. This reality created the force that shapes our society - the Will to Meaning, to make sense of all of this.

That force, that “lacking” - created the ancient and medieval religions, as well as the modern ones. From the greatest empires and conquests, to the most sacred household and family connections. We came up with stories, institutions, ceremonies - all to fill the chasm of our reality - of being on a floating rock in the sea of nothingness and darkness.

Viktor Frankl, the author of “Man’s search for meaning” described the “Will to Meaning”, from the despair and existential vacuum he observed in his patients and in his own experiences, human beings crave a sense of purpose, a reason to live. If that sense of meaning is absent, psychological unrest or “existential frustration” follows.

The Will to Meaning is not a poetic idea - it is a civilizational technology. Shared meaning lets strangers trust each other, coordinate, sacrifice, and build institutions that outlive any individual. Religion did this with cosmology, law, ritual, belonging, and founding myths. Meaning turns this vacuum, this chaos, into obligation. It tells you what is worth suffering for, and who you are while suffering.

What does it have to do with Activism?

Once you grasp this raw force and aggregate it across a population, the chaos of the world makes much more sense. The ‘crazy’ and the ‘great’ alike stem from the same engine: the desperate, underlying drive to make sense of the void. But because this force is so potent, it is prone to extraction. The same machinery that builds communities can be engineered for obedience. The same hunger that raises cathedrals can be redirected into movements that prize loyalty over truth. Meaning is fuel, and systems naturally learn to run on it.

Activism is the form of extracting that raw force, and applying it to a collective blurb. Instead of filling the gap in one’s existence with an individualistic pursuit of meaning, one’s filling that gap with an external one. A lot falls into that - including religions, patriotism, environmentalism, Queers for Palestine, etc.

While not necessarily a bad thing, it is often being harvested by others, powerful minds, to steer society.

It is okay to choose a mission, an ideal to contribute your meaning-seeking force to, but do it willingly and consciously - because society’s direction is at stake here - and so we might as well live in a better world.

— Adam

Every time a new medium emerges, it first mimics the old one. Then it finds its true form.

Personal computers started with file systems that mimicked physical filing cabinets: folders, directories, desktops. The internet’s early websites mimicked newspapers and magazines, complete with columns and mastheads. We didn’t yet understand what these mediums wanted to be, and what great new use cases they could enable.

Then something clicked. PCs evolved beyond digital filing cabinets into platforms for creation and computation - instant calculations across thousands of cells, pixel-by-pixel image manipulation, algorithmic music sequencing, database queries in milliseconds. The web broke free from its newspaper skin and became hyperlinked networks where anyone could publish, real-time messaging across continents, collaborative editing of shared documents, and streaming video on demand, replacing scheduled broadcasts.

When smartphones arrived, they initially just ran watered-down versions of desktop software. But the medium demanded something new. Touch-first interfaces. Location-aware services. Apps designed for your pocket, not your desk. The phone became the native device for the internet era, rewiring how humans communicate, organize, and access information.

Now we’re here again. AI is the new medium, and similar to how smartphones were the native device for the internet, we’re about to see AI’s native devices emerge. Everything signals the wave: Meta buying Limitless and now Manus, OpenAI’s upcoming family of devices with Jony Ive, and others racing to claim this space. The first explorers into a new interface don’t always get it right (Rabbit, Humane now, MySpace, Atari, Palm Pilot from the past), but that’s the point. The wave has started.

This comes out of the need for freedom to build freely with the medium, dance with its possibilities - unconstrained by the old paradigms and its ways. And this is exciting because the choices some of us make today will shape society for the upcoming decades. How we build this technology, what forms of communications and data manipulations are possible, how interfaces appear and disappear, who owns what, what values we embed into it: these decisions matter, and we have a white-clean canvas to paint our vision.

This is it. This is our moment.

We’re building Secondly because the next great consumer company isn’t a new app. It’s a new presence.

Today, AI is brilliant but episodic. You open a chat, you explain yourself, you get an answer, and the moment ends. It’s like having a genius friend who gets amnesia every time you walk out of the room. This isn’t because AI is limited. It is because we’re forcing this new medium into an old container called the session. We treat AI like a website you visit or a tool you query.

But if AI is a new medium, its native form won’t be a better chatbot. It will be continuity.

Secondly flips the model. Your AI stops being a destination you visit and becomes a constant that lives with you. It is not trapped inside one app or one browser tab. It flows across every surface you use, from your phone and laptop to your messages and the physical world around you.

Wherever you are, it is there. It sees enough to be relevant and remembers enough to be personal. It doesn’t wait for a prompt because it carries your context forward. It knows what you’re working on, who you’re talking to, and exactly what you meant when you said “the thing.”

When you bridge that gap, something fundamental changes. Your AI stops being software you use and becomes an extension of you. It becomes a second self that knows your world and can act inside it.

To actualize our vision for the future, we build upon three pillars.

Three Pillars: Context, Ownership & Hardware

Pillar 1: Context - Your AI is brilliant but blind

Today’s AI models are insanely smart. Some say they’re approaching PhD-level intelligence. But intelligence without context is useless.

Imagine running into Einstein on the street. Ask him to explain quantum mechanics or derive the theory of relativity, and he’ll blow your mind. But ask him about your life, your projects, your relationships, and he can’t help you. He doesn’t know anything about your life or your circumstances, so despite his genius, he’s useless to you at that moment.

That’s exactly where we are with AI today.

Quite often, when you interact with current AI products, you have to explain and re-explain everything. Long prompts, desperate context-setting, starting from zero. And even then, you might not get everything right in there.

Your AI doesn’t know about the conspiracy theory rabbit hole you went down last night at 2am. It doesn’t know about the inside jokes and pranks you like to play with your best friend. It doesn’t know that when you say ‘the thing,’ you mean that project you’ve been obsessing over for weeks. You spend 4-7 hours a day pouring your life into your phone, yet your AI is completely blind to all of it.

Intelligence is a commodity. Memory across your entire existence is the real thing.

Pillar 2: Ownership - Own Your Second Self

As Artificial Intelligence advances and gains full context of your life, it will become indistinguishable from you. It will think like you, act on your behalf, know things about you that no one else knows: not your partner, not your closest friends. Your secrets, your fears, your ambitions, the thoughts you’ve never spoken aloud.

This raises the question: who owns your second self?

Right now, we’re careening toward a dystopian future where OpenAI, Google, Meta (the big tech companies) own both the models AND your personal data. You’ll be “renting” your AI, never truly owning it. Always at the mercy of their terms, their policies, their business models.

And this is not a “maybe” future. It is the reality already.

For the past 20 years, we’ve been living in what we call the “Privacy Wild West.” We gave away our data to corporations without thinking. We handed over everything (photos, conversations, memories) to companies that could read it, use it, or hand it over with a legal request, not realizing what we were surrendering.

In 100 years, our belief is that people will look back at this era and ask: “How did they live like that? How did they give corporations the keys to the most sensitive information in their lives?”

And this is our cultural bet - that there’s another path. A future where you own both the inference and your data. Where your AI is truly yours.

This is civilization-level critical, but this is also not altruism on our end - on the contrary, we believe this is one of the biggest consumer plays one company can make. With the upcoming shift - ownership will matter more and more. This is no longer about privacy, it’s about property, about freedom of self.

Owning your second self, your personal AI - is a must.

Pillar 3: Hardware - Designing a New Operating System

Simply put, we are physical human beings, living in the physical world. Thinking you can build an omnipresent entity in a person’s life without building hardware is delusional.

Forcing ourselves to build software for old hardware is like trying to build Uber in a world of personal computers, or Twitter in the era of printed press. It doesn’t work, the medium just doesn’t enable it - and it misses the point. Alan Kay’s “People who are really serious about software should make their own hardware.” is more relevant than it has ever been.

The phone was designed for the internet era. Touch screens, app grids, notifications: all built for a world of manual human interaction. It wasn’t designed for background agents working extra hours for you, doing things on your behalf, delivering value while you sleep. But AI is a fundamentally different medium. It doesn’t need you to tap through menus or swipe between apps. It needs to see, understand, and act.

So we’re building from first principles. Hardware to software, bottom up. A device that embraces what’s possible with AI as the foundation. Always-on vision. Proactive intelligence. Action-taking capability.

That will incorporate a new operating system for your intelligence. An OS for your extension of self that lives everywhere: phone, home, car, wearables, every surface you touch.

The memory layer. The personalization infrastructure. The embedded device that lets AI actually co-exist with you in the physical world.

What We’re Building

Our vision is to build an omnipresent AI entity in every person’s life. That entity will exist across any surface you do, physical and digital as one.

It will build up a growing memory layer of YOU through observing your life. That means it will know you better than anyone else, be proactive, present, and relevant. And it will unlock entirely new forms of interaction we can’t even imagine right now, all built on top of this presence and memory operating system.

Through all of it, we fundamentally disagree with the previous way of building technology: rented, used by tech companies for profits. This is not altruism. It is a bet on the future market. We are convinced that the era of the user as a ‘product’ is ending. With the upcoming shift, ownership will matter more and more for individuals, like freedom of speech, property rights, etc.

The Hardware (Announced, In Production)

We started with a simple insight: the best way to give AI context about your life is to let it observe it. And where do you currently exist 4-7, even 8 hours a day? Your phone.

So - three months ago, we announced our hardware: a small device that attaches to your iPhone via USB-C, can see your screen and take actions on your phone - this allow your AI to continually observe and build the memory of you based on your digital usage - who are you talking to, what things you’re interested right now, etc. The response broke the internet.

Now we’re working on production. Meeting with factories around the world, testing plastics and textures, doing durability tests. Building a production-quality device that will ship with a mobile app and bring your AI to your phone.

We might decide to ship a dev kit sooner, but regardless - we’re not waiting to start delivering value.

The Mac App & Memory OS (Launching Soon)

While we are working on getting the hardware produced, we will be releasing our Mac app first. This is in part align with our local-first approach, an app that will later also serve as your backend server, and because on Mac, permissions are more open and we can start building your second self today. We can deploy our multimodality-based memory system: taking inputs from your screen, audio, text, and files to build a complete understanding of your digital life. A 24/7 timeline of everything you do. Your second brain, always learning, always there.

Offline & The Private AI Cloud

We believe everything should be able to run locally if you choose to. Our backend will be distributed to you to run on your home computer or server - but we are also aware of the great benefits of the cloud (not needing good hardware at home, for example), hence we are working hard to ship a private cloud architecture - where we provide you top GPU performances, backups, availability - while keeping your data encrypted, so even we can’t see it.

The ownership remains yours. You can always take your data home, run the backend yourself on your own computer, without needing us at all.

We will share more about our architecture & deployment very soon.

How It All Comes Together

Your brain currently has 50 tabs open. You’re trying to remember to text your mom back, wondering if you ever sent that file to Mark, and holding onto a vague reminder to buy flights for October. These are “open threads”, background processes that drain your energy.

Secondly closes those loops.

It tracks the context that usually slips through the cracks. When you open a chat with a friend you haven’t spoken to in months, Secondly whispers the context you’ve forgotten: “You promised to send him that article about architecture last time,” or “Ask how his surgery went.” It turns a cold interaction into a warm, continuous connection.

It holds your mental “to-do” list so you don’t have to. When you mention in passing, “I need to get a gift for the anniversary,” Secondly logs it. It doesn’t nag you; it waits until you’re browsing online or, in the future, near a store to nudge you. It manages the loose ends of your existence.

We’re experimenting with dynamic UIs, widgets, and extended surfaces that let these interactions flow naturally. We’re working on a memory paper to share our approaches to multimodality-based memory. Research we’ll be sharing with the world soon.

This is just the beginning. The interactions that’ll be possible soon are crazy, it will feel like magic - and we hope to bring y’all along for the ride.

Your second self, your Secondly.

— Secondly Team

P.S. - If you want to be among the first to get your hands on Secondly, join our waitlist at Secondly.com

Reading the rich history of that period, particularly the fall of the Roman Empire and its impact on the world for so many generations, makes you think differently about the world we live in: nothing is permanent. Everything is consistently changing - and sometimes, you need to look at a larger scale to see the big shift happening. Rome didn’t fall in a day but declined for generations, and so is the current world we all take for granted.

In this essay, I will explore how our world is changing, the powers that drive that change, and what might be ahead of us as a society, with the challenges we are already facing when going through these early days of the shift. Tech companies like Apple, Worldcoin, OpenAI, SpaceX, and others, including some individuals, are challenging everything we know about our world and institutions. In particular, the geopolitical actors involved with the social contract between citizens and governments.

Some of the changes are positive, some are less so. To start our journey, let’s first get some context for the world we are living in: The Modern State (Western Edition).

The Modern State (Western Edition) in a Nutshell

The political system we live under today, the modern state, is a relatively new creation. It’s a framework that claims full control over a defined territory, runs professional armies and tax systems, governs through law rather than whim, and is bound (at least on paper) by constitutions and elected institutions. This is the blueprint we’ve come to accept as “normal,” but it didn’t exist in this form for most of human history.

This blueprint was shaped by the ideas of European thinkers from the 17th and 18th centuries - Thomas Hobbes, John Locke, Montesquieu, Jean-Jacques Rousseau, and others. These were not isolated philosophers sitting in ivory towers. They were responding directly to the turbulence of their time: religious wars, collapsing feudal orders, early capitalism, rising empires, and technological change. They revived and reworked much older ideas, from Roman law, canon law, medieval city-republics, and natural law philosophy, and adapted them for a world that was entering modernity.

Hobbes tried to make sense of the need for order after civil war, proposing a strong sovereign to secure peace. Locke argued that governments should protect natural rights like life, liberty, and property, and that their legitimacy comes from the consent of the governed. Montesquieu emphasized the need to divide power between different branches to prevent tyranny. Rousseau focused on the collective will of the people and how sovereignty belongs, ultimately, to them.

These ideas came together as a political operating system, and they had enormous influence on the next wave of change: the American and French Revolutions. The U.S. founding fathers borrowed directly from these thinkers to draft the Declaration of Independence, the Constitution, and the early institutions of democratic governance.

Again, those ideas didn’t appear out of nowhere. The ground had been shifting for centuries.

The Military Revolution in early modern Europe gave kings and states new powers - standing armies, gunpowder weapons, and the ability to crush feudal lords. To pay for it all, they had to build tax systems and centralized bureaucracies. The printing press revolutionized the flow of information, creating a public that could read, argue, and eventually imagine itself as a shared political community. The Reformation shattered the Church’s monopoly on truth and authority, opening up space for secular political theory. And the rise of mercantile empires and colonial expansion forced states to become more organized, standardized, and powerful in order to compete on the world stage.

The Peace of Westphalia in 1648 is often marked as a turning point. It formalized the idea that each state had ultimate authority within its borders, sovereignty, not shared with popes, empires, or supranational bodies. This model of the state, based on borders and autonomy, would become the dominant one in Europe and eventually spread around the world.

By the late 1700s, it all came to a head. The revolutions in America (1776) and France (1789) put into practice the theories of Hobbes, Locke, Montesquieu, and Rousseau - combined with the machinery of state power that had been built up over the previous two centuries. They didn’t create the modern state out of thin air, but they added a new element: the idea that the state must be legitimated by the people. That it needs a social contract.

Before the rise of modern nation-states, power took many forms: feudal kingdoms, empires, sultanates, theocracies, city-states, leagues of merchants, and even chartered corporations with armies. The British East India Company is a clear example, governing large parts of India with its own military and tax apparatus until the Crown took direct control in 1858.

Social-contract thinkers asked a basic question: why do people obey governments? Hobbes, Locke, and Rousseau offered different answers, but all shifted legitimacy away from divine right toward consent, whether explicit or implied, between rulers and ruled. People give up some freedoms in exchange for protection, order, and the preservation of their rights. When that bargain is broken, Locke and, in a different register, Rousseau allow that the people may change the arrangement.

That concept may sound obvious today, but it was radical at the time. And it still matters. Because if the foundation of our governments is a contract, and if we’ve seen contracts break and be rewritten throughout history, then we have to accept that our current system, the democratic nation-state, is not the endpoint. It’s a chapter.

The Shaping of a New World

I argue that there is a power and influence transition from “traditional” institutions, like governments and central banks, towards a for-profit, commercial entities. This transition is mainly led by three factors:

1. Capitalism, and the untapped market of digital protection.

2. Decreasing confidence and trust in our institutions

3. Modern challenges that are not addressed by current institutions and hence demand new approaches to govern

Let me explain:

I. The Untapped Market of Digital Protection

Governments originally formed to offer structure and protection: managing resources, settling disputes, defending against outside threats. But in the last 20 years, a major shift happened: our lives moved online. And governments haven’t followed.

In the digital realm, the idea of “law and order” becomes blurry. If someone breaks into your house, you call the police. But if someone breaks into your inbox, or hijacks your digital identity, it’s unclear who protects you. There’s no sheriff in the metaverse.

But most Apple users would probably say: “Apple does.”

Whether explicitly or not, people have begun outsourcing their sense of digital protection to private companies. Their “digital social contract” is no longer with the state - it’s with Apple, Google, or OpenAI.

As I wrote in Proof-of-Personhood: Tech’s Triumph Over State Monopoly, the last monopoly many states still hold is on identity. But even that is fading. Apple, through its devices, biometric hardware, and software integrations, already has a more reliable sense of who you are than most governments do. And as AI evolves, reliable identity - knowing who is human - becomes one of the most valuable pieces of infrastructure in the world.

II. The Rise of Private Geopolitics

But we’re not just seeing this shift online. It’s already playing out on the world stage.

When the Ukraine-Russia war began, SpaceX became more geopolitically relevant than most NATO members. Starlink provided internet to Ukrainian troops at the front line - something no nation-state could offer at that scale, at that speed. And Elon Musk, a private citizen, ended up influencing the outcome of a land war in Europe.

This is not a one-off. Private companies are now negotiating at the highest levels of diplomacy, and some are even moving toward nuclear capabilities (tweet and tweet)

III. Decreasing confidence and trust in our institutions:

When Julius Caesar led the revolution that transformed the Roman Republic into the Roman Empire, he did so in the name of the people, fighting against the corruption and elitism of the Roman Senate, which had become dominated by a small group of aristocratic families. This is one example out of many throughout history of how growing distrust in institutions and the ruling powers - eventually breaks down into revolution, which leads to a shift in power. It happened with the Peasants’ Revolt of 1381 in medieval England, the American and French revolutions, and the modern-day Arab Spring - all stemmed from growing distrust in the ruling powers.

Today, and in the past few decades, we are seeing decreasing confidence and trust in our institutions. And in the case of Democracy it’s even more significant - as it is based on trust. A trust that is given by the people, the governed, to the government. If trust does not exist, then the social contract starts to weaken, and the government’s legitimacy becomes more in question. Then revolutionaries ideas and movements break the tension, and the power shifts elsewhere.

IV. Modern Challenges Need Modern Governance

Our existing institutions - governments, legal systems, education systems - were built for a world that moved slowly. A world where change was generational, where policies could be debated over years, and where society had time to catch up.

That world is gone.

Today, AI improves weekly, biology is programmable, and robots are learning to walk, talk, and reason. Every month brings a new capability and a new risk, while the institutions meant to govern all this are still operating in a 20th‑century rhythm, inside 19th‑century buildings, using 18th‑century ideas. They were not built for this speed.

We spent years debating cookie‑consent banners while entire surveillance ecosystems grew around us. The European Union spent a decade negotiating how to “protect” users online, and the result is that we all just click “Accept All.” Meanwhile, companies are building models that understand our speech, faces, and emotions better than our governments ever will.

It is not that public institutions are lazy. They are using the wrong tools. They were designed to regulate land, labor, and factories, not data, algorithms, and synthetic life. The legal frameworks that once managed ships and steel are now trying to make sense of neural networks and DNA printers.

Education shows the same mismatch. Schools were created for the industrial age: rows of desks, bells, and memorization. That system worked when stability was the goal. Today, everything changes faster than curricula. Students can learn more from YouTube, Discord, and AI tutors than from the classroom, given they have enough agency to pursue it. The factory model of learning simply cannot keep up with the exponential curve of knowledge, and these are not the humans we are in demand for.

Commoditizing The Social Contract

“As Huizinga said of Europeans in the late 15th century ‘Their whole system of ideas was permeated by the fiction that chivalry ruled the world.’ This has a close second in the contemporary assumption that it is ruled by votes and popularity contests. […], the idea that the course of history is determined by democratic tallies of wishes is every bit as silly as the Medival notion that it is determined by an elaborated code of manners called chivalry” (Book “The Sovereign Individual”, Page 99)

What I see happening, is that due to the reasons mentioned above, the social contract that defined our fabric of reality in the past 250 years, is moving from governments to for-profit institutions. Why? Simply because they’re able to innovate and stay relevant in these turmoils times.

This shift presents both opportunities and challenges. On the positive side, commoditizing aspects of the social contract could drive innovation in areas previously neglected by traditional governance. Digital protection, identity verification, and rapid response to technological challenges might finally receive the attention and resources they deserve. Market forces could push both public and private sectors toward better service delivery, creating healthy competition.

The emergence of multiple players in what was once exclusively government territory could lead to more resilient systems. When SpaceX provided internet to Ukrainian forces, it demonstrated how private capabilities can complement, or sometimes exceed, traditional state capacity. This redundancy might actually strengthen our collective resilience rather than weaken it.

However, this transition requires careful navigation. As more civic functions shift toward profit-driven entities, we need robust mechanisms to ensure these new arrangements serve human flourishing rather than just shareholder returns. The key question isn’t whether this shift will happen, it’s already underway, but how we can shape it to preserve the values we hold most dear.

I’m not particularly attached to democracy as a system per se, but I am deeply committed to the Western values it has helped protect: freedom of speech, individual liberty, property rights, scientific inquiry, artistic expression, and the right to dissent and dream. These principles have enabled centuries of human progress, and they’re worth preserving regardless of the institutional framework that carries them forward.

The companies leading this transition still need us: our buying power, our participation, our data, our trust. This dependency is our leverage. They must still compete for our attention and allegiance, which means they must still serve our interests, at least for now.

But this window won’t last forever. As these entities become more essential to our daily lives, as their capabilities exceed what traditional institutions can offer, the balance shifts.

The shift is happening whether we acknowledge it or not. We can engage thoughtfully with this transformation, ensuring that whatever emerges next serves human flourishing, or we can let it unfold without our input.

- Adam Cohen Hillel

Polymaths, individuals who master multiple fields at the highest levels, were much more common in past eras. Think of the founding fathers of the United States, Benjamin Franklin and Thomas Jefferson, who were inventors, writers, diplomats, and scientists. Or maybe the most famous of all, the man of the Renaissance, Leonardo da Vinci, who was a painter, engineer, and anatomist. They all mastered various fields and made great contributions to each of them individually, as well as through the combination and crossing of knowledge.

The most recent age of polymaths lasted from the Renaissance, through the enlightenment and early modernity, up to the first half of the 20th century. Then, as knowledge deepened rapidly due to the exponential growth of scientific and technical fields, most notable post-World War II, it required more specialization, and so it became nearly impossible for one person to be well-acquainted in multiple disciplines in a fruitful way. The sheer volume of information and the complexity of fields like physics, mathematics, biology or even philosophy required intense focus, making true polymathy rare (though there are exceptions).

That now changes, as we enter an intelligence explosion fueled by AI. A shift that gives us immense leverage to learn and experiment across disciplines. In the hands of the curious, the ambitious, the high-agency few, this will bring back the age of the polymath. The fluidity of ideas, curiosity-led tinkering, invention, rethinking, discovery — all the magic of the Renaissance and the Enlightenment — is ahead of us. It’s available. But don’t get confused: it won’t come easy.

We’ll see individuals contribute across multiple fields, pulling ideas from one into another. Culture, the arts, politics, economic frameworks, engineering, biology — they’ll all become one big game field. So will new frontiers: engineering breakthroughs in space, and revivals in older fields like manufacturing. The doors are open, the tools are ready — and it’s all up for the taking.

The internet was an early teaser of that — making knowledge much more accessible, and free for all. You can quite literally learn almost anything humanity has to offer on the internet, though mastering it still requires many many hours of learning, experimenting, failing, debugging.

It will be so with AI too, but the speed of iteration and the depths one can get into will be dramatically different. And the thing is — I don’t have to master deeply any field, I mostly need to be able to speak its language and play in a sandbox, a simulated environment, to test my ideas and assumptions. That is how I learn best (by doing) - and AI will allow us to do it across many fields, not only in computer science.

As the barriers to knowledge, information processing and experimentation are reduced — the real value won’t come from staying within one silo. It will come from fluidity — the ability to take ideas from one domain and apply them in another, to see connections others miss, to invent at the intersection. Modern breakthroughs aren’t isolated. AI will reshape the economy, which will reshape culture, which will reshape politics and institutions. Biology and computation are merging. Brain-computer interfaces will demand insights from neuroscience, hardware engineering, ethics, and law. If you treat each domain as its own fortress, you’ll miss the bigger shift. But if you’re fluid — if you can roam, connect, remix — you’ll be at the frontier.

I don’t have a “formal” high education myself, though I’d place myself at the top percentile of computer scientists. I spent, spend and will spend an order of magnitude more time learning things compared to others who specialized in it formally. I had spent countless hours on the internet trying to figure things out and brute-force myself into understanding. I believe that with AI I can do it again and again, ask endless questions, have a tutor that can just explain until I get it - and become a masterchef, a polymath, in many fields and professions.

I don’t think I’ll become the best in any one field — and definitely not in all of them — by taking this path. But I believe there’s immense value and leverage in choosing polymathy over specialization. That said, I might be wrong. There are still many unknowns. Will humans even stay relevant? And if we do, maybe the real leverage will lie in going deeper, not broader. A few friends I spoke with recently had the opposite take — this is the age of ultra-specialization. If AI can take you 80% of the way in any domain, then the most valuable thing a human can do is push into that last, difficult 20%. I think about that a lot.

In the age of AI, bottom-up, brute-forcing learning and taking action will lead to the best outcomes.

I’d argue that my informal education has put me in a better position to take full advantage of this trend — because I’m tuned to self-learning, to exploring, to asking questions, to “try and see if I can do that,” rather than top-down feeding of knowledge. I’m continuously learning. I’m not tied to a four-year block for learning. I’m on an endless path of exploring, of trying, of testing things out. With AI, I can do it better, learn faster, dive deeper, test more. I get curious about a specific topic, and then I’ll spend weeks, months, sometimes years — on and off — diving in. I did that with computer science, code, and machine learning. I’m doing it now with history, philosophy, and electrical engineering. I hope to continue learning that, and more: mechanical and aerospace engineering, economics, biology.

So to finish up — I’m extremely optimistic about what’s ahead. I believe AI will enable us to revive the old idea of mastery, of polymathy. By reducing the barriers to knowledge, and empowering the ones who have agency and curiosity, we’ll see a wave of invention, discovery, and new thinking across culture, science, and society. AI is ours for the leverage. The time is now — go learn, go deep, never stop. And please — produce. You are precious because you are agentful and curious.

Adam Cohen Hillel,
Aspiring Polymath

From the founding fathers of the United States during the Age of Revolution, to the early Zionists who built the modern State of Israel in the age of rising Nationalism, to the pioneers of the Information Age who gave us personal computers and the internet, to the explorers who pushed the boundaries of the known world during the Age of Exploration—yes, they were all great. But they were also products of their time. Their greatness came from pushing the frontier of their era.

Put simply—I can’t discover America, or sail to an unknown world—it’s already been discovered and mapped (or has it!?). I can’t reestablish the State of Israel and revive Jewish national aspirations—as I was lucky enough to be born into it, after nearly 2,000 years of exile, persecution, and generational longing.

Yes, sometimes I wish I was born in a different time. Boarding a ship to the New World with Columbus (or being Columbus myself), or establishing a colony in some uncharted territory. Or the feeling of touching ground in the promised land after hearing my parents and grandparents talk about Jerusalem every single week—and helping build a country from scratch, a state for the Jewish people.

But that world is gone. And those struggles are not my generation’s struggles.

These men weren’t great in a vacuum. They became great because they pushed the frontier of their time forward—within the unique cultural, political, environmental, and technological conditions that allowed them to do what they did.

That’s not to say there’s no room for agency, or that the trajectory of humanity is guaranteed regardless of these men. Not at all. The Dark Ages are proof of the possibility of stagnation. What I’m saying is—it’s all deeply intertwined. Possibility is bound to context.

It’s somewhat similar in business and innovation. Sometimes you’re right—but it doesn’t matter if you’re too early. History os full of people who were decades or centuries ahead of their time—whose potential only got recognized long after they were gone. Ada Lovelace, Alan Turing, Emily Dickinson, Vincent van Gogh, and many others.

That moment when the “stars align” (culturally, politically, environmentally, technologically)—that’s what I call a Wave of Greatness. A new frontier opens up, and those ready for it can ride the wave, push boundaries, redefine how we conceive society, and become one of the Greats.

And if I want to become great, and I do, I need to understand the relevance of my generation, of this moment. I need to be ready to catch the wave, which might also never come in my lifetime. But even if it doesn’t come in my lifetime, you can’t know until your last day, I’ll need to push to discover it regardless and create the conditions for it to appear—whether for myself, or for the next generation.

For the American Revolution, what benefited Jefferson and Adams was their fluency in law, writing, and the press, and Washington was years of military experience. In the Early Information Age, it was Computer Science and software that provided the most leverage to who knew how to wield it.

You have to look around and ask: what is coming, what wave should I prepare for? And I think, for our time, there are a few areas that might become relevant, for me at least - to have a chance at greatness. I probably miss a few things, but the point is to stay tuned.

Space Exploration and Colonizing Mars — I believe that in my lifetime, there’s a real possibility we become a multi-planetary species, mostly thanks to Elon Musk. That means we’re in the early days of yet another Age of Exploration—2.0 (aka Space Exploration). Which means my aspiration to explore uncharted territory might actually materialize. And those who do it—those who go first—might be remembered like the greats: like Christopher Columbus and the people of his time, but in space.

So that leads me to think: when will that happen? How can I best be prepared to be part of it, if I want to be? becoming a doctor, an engineer, or something else entirely? inventing space-internet? or just having enough courage and resources is enough?

The Age of Artificial Leverage — With AI, we might need to rethink almost every part of our society. Relationships, institutions, education, economics, work, science—everything.

I deliberately used the word leverage here because I believe AI, combined with other forces, could trigger a renaissance of ideas, discoveries (scientific, engineering, etc.). I believe AI brings the threshold back down—it makes the world full of “low-hanging fruits” again. For discovery. For invention.

For example, I’ve been thinking a lot lately about interplanetary internet connectivity—a future problem that will absolutely need solving. I’m not the most skilled person for the task, but I’m also not entirely disconnected from it. So I went into a few-hour-long conversation with Grok, and came up with some interesting ideas about what could be built—from fast replicas, to a new kind of internet that’s predictive. I called it: Dynamic Predictive Interplanetary Network (DPIN).

The details of the idea don’t really matter—the point is the leverage we now have. The ability to discover new things. To invent new things. At a time when new things need to be invented—not just technologically, but culturally too (see next).

Big Shifts in World Order — I’m convinced the industrial-era social contract and the political thinking that came out of the late-18th Century will become somewhat irrelevant. Nation-states still wave flags, but money, code, and ideas jump borders instantly. The internet economy, especially with AI, moves faster than any parliament can legislate. I don’t know what will come after, but I believe we’ll need to rethink it all—John Locke the foundations of the modern nation-state, democracy, law, etc.—and fit them to this new age. And if I’m completely honest, even without AI and space colonization, this would’ve been needed—we’re still defined by industrial-era political thinking, and we haven’t yet adjusted our institutions to the internet. That’s also why I believe the internet is very much in its EARLY DAYS. I don’t know what that will be—Balaji’s network states? DAOs? Fluid citizenship? Data rights? Privacy? Software-led decision-making and legislation?

The Jewish Israeli Frontier — Sadly, I have a growing sense that the Jewish people are approaching a major turning point—in politics, in religion, in identity.

In Israel, there’s a deepening split—between left and right politically, and between Orthodox and secular Jews religiously. I also fear that the growing “political” anti-Israel sentiment around the world (which, in my view, is just rebranded antisemitism) might create a split between Israeli Jews and Diaspora Jews. A split that, until now, hasn’t really existed—both groups have mostly been aligned and cooperative since the very beginning.

This wave, if it continues, might be devastating. It may require a great unifying force—something strong enough to pull Jewish people and culture together in a time of fragmentation. It might demand rethinking our political institutions. Maybe even our religious ones.

All of the above are possibilities, and I need to position myself to be ready for them. In practice, I think I need to:

• Study all aspects of knowledge. From history, ideologies, and major events, to classical literature and thinkers, to modern ones. Economics, psychology, systems thinking—all the way to religion, practice, and lived experience.

• Accumulate enough financial resources to fund my ambitions and experiments. To have the freedom to think independently. To go early on space exploration if the opportunity comes, or to pursue political aspirations if that's the path.

• Build a name for myself across the disciplines—as a thinker (hence writing publicly), business (hence building a company, becoming financially successful), and technology / innovation (both). Position myself as someone who matters is important. George Washington became the first president because he was positioned well as a great general beforehand — and when the time was right, and the wave hit, he was ready to become one of the Greats.

Future Human Generations, Let’s Go!
Adam Cohen Hillel,
June 7th, 2025 — a random Denny’s Diner in California

*As I live in times of great sensitivity, I feel the need to emphasize that by using the term Great Men, it also encompasses Great Women.

Last time I was as lost, slightly more even, was in 2020, when I spent a few months after finishing my service trying to build a new company, in the AI space. I locked myself in my room, all day, everyday - learning all about AI, and building, before it was sexy or cool. I was early to the LLM trend, and around that time, almost three years prior to ChatGPT release - I was already playing with GPT-2. With products like AIBro, which was an LLM helping you talking with girls, or Vivify, an app to clone your beloved ones from text messages, which I built after my grandma passed away. But I couldn’t get it off the ground, and slowly slowly I got lost.

I lost my way and broke all my promises, and joined a startup. It had its benefits, and I am grateful for that time - but it distracted me from my true quest.

I want to be the top innovator of our time. I want to contribute greatly to humanity, and our collective story, as well as to the Judaism-Israelites story. I want affect people’s everyday life, and be a character of aspiration and inspiration. I want to be a lover, a father, with a loving family. I want to be a better friend, the supporter kind, the one that is rooting for you. I want to be more easy-going, more friendly, better with people. I want to be one of the greats. I want to be the greatest. I want to matter.

I have ambition, I have ideas, and I execute. I don’t shy to express myself, and I put my creations out there in the world. And yet - I don’t feel like I am any close to the person I want, and ought to be — and every day, every year that passes, I feel ever farther from him.

I believe the reason for this is that I am missing one core feeling—the sense of a calling. Everything else is in place, and I was put here with my character in the perfect way to receive a calling. Yet, my faith is falling short, and I see the world as more meaningless than most.

It is as if I was gifted with too much agency, it all feels tasteless. I only wish I had faith, I had known my purpose. The thin thread that is still holding it all together is of the “soon”. I wonder if it will hold forever.

I do have a storyline in mind about how things can positively fall into place after all. I need to build a very successful company in the next two years while also increasing my public presence with my authentic self. I need to put myself out there not just with my products, but with my thoughts, ideas, and existence. I can’t keep this side-notes on Substack—I need to put it out there. That includes ideas around my shows, judaism, philosophy, history, society, startups, companies, and the future of humanity. I need to get more involved with more startups, from investments to building ones. But more than anything: I need more fucking wins.

I need to think bigger, I need to work harder.

The time passes, and I am losing momentum.

How can I become a main character?

Who am I?

What is my story?

—
This very essay and others I write here on Substack are actually not the first time I am writing about my struggles and my challenges of my journey. I started this concept in the 2018, with similar blog posts about my ambition and journey of building something big and meaningful — I encourage you to find them! ;)

There was one small exception—a reality show that caught my attention a bit more: Shark Tank. I thought watching entrepreneurs pitch their ideas was cool at first. But as I kept watching, I realized that Shark Tank only scratches the surface. Sure, there’s that burst of adrenaline during a pitch, but it never shows the messy, day-to-day grind behind building something great (something I experienced myself later on). The format offers quick, high-pressure moments, but it misses out on the depth of what really goes into creating a startup, and not because it is any less interesting and a good material for a tv show, no.

Then I discovered the HBO sitcom Silicon Valley. Now that was different. This show follows a group of founders living together in what called a “hacker house” (we will get to that in a moment), and it captures something real about the journey of starting a company. It isn’t just a series of dramatic pitches; it’s about the daily struggles and the little moments of building a company, in an intense, grinding house full of silicon valley ambitious minds. Those moments can be hilarious, emotionally engaging, and deeply involving. The day-to-day life it portrays isn’t just filler—it’s full of potential to be both entertaining and genuinely valuable for anyone watching.

Last year, almost out of nowhere, I decided to pack my bags and head to San Francisco. Before I knew it, I was living in a real-life Hacker House—a place that felt straight out of the sitcom I’d loved growing up. Suddenly, I wasn’t just watching a show about scrappy founders; I was one of them, sharing a house with more than ten other ambitious people, all working on their own startups. The late-night brainstorming sessions, the frantic pitch rehearsals, the endless working hours and the all-nighters before big launches—it was chaotic, intense, and exactly the kind of energy I’d always imagined.

As I settled into this new reality, I couldn’t help but think back to my teenage years in Israel. While my classmates were glued to reality TV shows full of gossip and drama, I was pulling solo all-nighters, trying to bring my big ideas to life. I longed for a culture that genuinely celebrated hustle—a place where being ambitious wouldn’t make me the odd one out. Suddenly, living in a Hacker House, it struck me that showcasing this environment in a reality show could spark the cultural shift I’d always hoped for. Instead of polished, 90-second pitches like on Shark Tank—or worse, shows that lock people in a house just to watch them fight—it would shine a light on the unfiltered, messy, and often hilarious process of actually building something from scratch. And maybe, just maybe, it would inspire the next kid who’s tired of feeling alone in their ambition.

And that’s how The Hacker House Show idea was born.

I want to blend the raw entrepreneurial spirit of Shark Tank with the authentic, day-to-day storytelling of Silicon Valley—not just for the founders who “get it,” but for anyone who’s ever felt that spark to create something amazing. Maybe this show will inspire the next young dreamer—someone sitting in Tel Aviv, Tokyo, or anywhere else—who just want someone to counter or even over-counter their own ambition. They’ll see it’s possible, it’s worth it, and most of all, they’re not alone in the craziness of chasing an idea that just might change everything.

San Francisco, 2024:

Back Story

Simply put, the modern term “democracy” refers to a system of government where the power of the state lies with its people, who govern indirectly through elected representatives.

—

To put it even more simply, citizens vote every few years to elect politicians who will manage the government on their behalf, until the next election.

—

By “managing the government,” we refer to organizing and directing the state’s resources, laws, and policies to (ideally) serve its people.

—

Elected representatives set priorities and policies, such as taxes, tariffs, and public programs. They allocate resources, like funding for the military or infrastructure, and work with lawmakers to create and enforce laws.

—

This system has its roots in the early democratic experiments of ancient Greece. However, most of the ideas shaping modern democracy are somewhat recent, dating back about 250 years to the Enlightenment era and the age of revolutions, including the American and French Revolutions.

—

The United States is considered one of the earliest modern democracies. Its democratic foundations began with the establishment of representative governance, the House of Burgesses, in Jamestown, Virginia, in 1619.

—

These ideas were solidified with the drafting of the U.S. Constitution in 1787, which created a constitutional system of government (i.e., a government based on written rules everyone agrees to follow).

—

It introduced principles like the separation of powers, a system of checks and balances, and a framework for protecting individual rights through amendments like the Bill of Rights.

—

That is in contrast to previous ways of governing, such as monarchies or feudal systems, where power was concentrated in the hands of a king, queen, or a small elite.

—

In those systems, ordinary citizens had little to no say in decision-making processes, and leadership was often hereditary. Modern democracy, by empowering the populace and institutionalizing rights and freedoms, represented a radical shift towards inclusivity and accountability in governance.

—

Technology and technological revolutions have consistently driven major shifts in governance.

—

The Gunpowder Revolution of the 14th century ended feudalism, as centralized monarchies, armed with professional standing armies, eclipsed the power of feudal lords.

—

The Printing Press in the 15th century spread ideas in local languages, fostering national identities and laying the groundwork for nation-states.

—

Advances in navigation and shipbuilding during the Age of Exploration enabled the European colonial empires of the 15th–19th centuries.

—

The Industrial Revolution reshaped societies by driving urbanization and creating the modern working and middle classes. Concentrated in cities, these groups organized to demand better conditions, fair wages, and political representation.

—

The emergence of the working middle classes not only challenged traditional power structures but also shaped the political and economic thinking that defined the 20th century - driving movements for democracy, labor rights, and social reforms that continue to influence governance today.

“We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness.--That to secure these rights, Governments are instituted among Men, deriving their just powers from the consent of the governed, --That whenever any Form of Government becomes destructive of these ends, it is the Right of the People to alter or to abolish it, and to institute new Government, laying its foundation on such principles and organizing its powers in such form, as to them shall seem most likely to effect their Safety and Happiness.” From Declaration of Independence, July 4, 1776

In between the past and the future

Although 250 years may not seem like a long time ago, the world was vastly different from today. At that time, the majority of the population worked in agriculture, a labor-intensive occupation that consumed most of their working hours. Communication relied on postal systems—there was no internet, telephone, or anything remotely similar—so access to information was slow and limited. Transportation was also extremely slow, with trains only beginning to emerge.

—

It is important to note this, because the democratic system created by the Founding Fathers of the United States—remarkable as it was—was designed for a world vastly different from ours, shaped and constrained by the tools, knowledge, and ways of thinking available at the time.

—

As we mentioned already - at its core, modern democracy is founded on the simple principle that “the power of the state lies with its people, and representatives govern on their behalf” (whether this always holds true is a topic for later debate).

—

This principle defines what we call an “indirect democracy,” where citizens do not govern directly but instead elect representatives to make decisions and manage the government on their behalf.

—

This principle, rather than being a defining feature of democracy, was more of a necessity born out of the technological and logistical constraints of the time.

—

Direct democracy—where citizens themselves vote on laws and policies—was simply impractical in a world where communication was slow, transportation was limited, and populations were dispersed over vast areas.

—

You could imagine the challenges: how would a nation of farmers and laborers, most of whom were tied to demanding daily work, come together to deliberate and vote on every issue? How would decisions be communicated effectively across such distances without the tools we take for granted today, like the internet or even telephones?

—

Representatives became the solution, not because it was an inherently superior system, but because it was the only feasible one.

”Democracy is the worst form of government – except for all the others that have been tried” — Churchill.

Issues

Democracy, with all its merit, is not perfect. It’s a living system that must adapt to the new challenges of the age. Especially in the face of yet another big technological shift - AI.

—

For one, the age-old problems of corruption and bad incentives still plague democratic institutions. While it’s difficult to bribe an entire population of millions, power often concentrates in the hands of a few decision-makers, making systems susceptible to manipulation. Influential interest groups, bureaucrats, and lobbyists - leaving ordinary citizens sidelined.

—

Our world has shifted from the 20th-century industrial era to the digital and information era of the 21st century, with AI accelerating this shift even further. Yet, most of our institutions were created for and shaped by the industrial era. The changes brought by our modern era demand a complete rethink of policies around education, labor, and innovation. But traditional, slow-moving governance structures struggle to keep up with rapid innovation.

—

Some changes need to happen now, not in a few months or years.

—

People and human nature are more rounded than modern democratic system allow them to be. They can agree on certain issues with one party or politician, and on a different issue, with another.

—

There will be issues people would care a lot about, as it affects their life directly - while those same issues won’t matter at all to others.

—

The four-year term in most modern governmental elections is an arbitrary number that may have acted as a feature to guard against the concentration of power, yet it has also had the unintended consequence of incentivizing short-term thinking, and populism.

—

We need more responsive and transparent governance. Indirect democracy, while serviceable in the past, will almost certainly fail to capture the true will of the people.

—

Hence, the proposed direct democracy (exercised through Open Source Software, which we will get to shortly)

—

Direct democracy, as the name suggests, occurs when citizens vote directly on governance matters. It is a form of government that is practiced today, but in limited ways. For example, Switzerland held several referendum in 2024, and the United Kingdom held one for Brexit. However, these remain relatively isolated instances of direct democracy.

—

Impracticality Critics

Critics of direct democracy argue it’s “impractical.” They claim everyday people can’t or won’t engage with every policy. Personally, I find that these old arguments often reflect elitism—assuming citizens can’t be trusted to govern themselves. It is as if they support “the nanny state”. A way of thinking that completely contradicts the ideals of the Founding Fathers of the United States, and the democratic innovation.

—

Today, with instant connectivity - and more importantly, with AI - tech reshape what’s possible.

—

With AI, complex legislation can be broken down into plain language or even short bullet points. If the state is operating using code (more about it shortly), then participating, with AI become accessible to all.

“We reject: kings, presidents, and voting. We believe in: rough consensus and running code” - David Clark, MIT researcher and internet pioneer

Three Building Blocks of the Future

Software:
Big part of our world is already run by software and automated algorithms. From our financial systems, households (keys, cars), our most used communication tools, to the way we learn, etc.

The reason our world is running on software, is because it is efficient and flexible. It allows fast iteration, and a scalable way of value creation. That’s why we saw a lot of innovation in “the world of bits” compared to the “world of atoms” as Peter Thiel like to say.

—

Open source software:

Open source software is a subset of software that refers to a program that its code is publicly shared and available online for people to see, run, and modify themselves.

Most of software is mediated to consumers through the internet, which mostly runs on open source software (Linux, Android).

In its technological context alone - the open source movement was one of the biggest saviors of the enlightenment ideas in recent decades. Enabling individuals freedoms, especially the freedom of speech, at times when it could have been restricted more easily.

It is the reason you can freely open a web browser and surf almost any website you desire, without paying or asking for permission.

However, this was not the only way the digital revolution could have played out. In fact, many companies (including Microsoft, AOL, etc.) tried to fight it.

Sadly, since the iPhone’s AppStore was released on July 10, 2008, the open internet has been at risk.

Big tech companies, like Apple and Facebook, have gradually taken over what made the internet open and permissionless, with their closed-garden products and gatekeeping. But that is a topic for a different essay.

—

Blockchain:
Blockchain is often described as a “decentralized database,” which essentially means we can record, store, and verify data without relying on a single authority, and without worrying about someone tampering with the data.

That database is spread across a network of computers (nodes) rather than a single server, using cryptographic links that make tampering nearly impossible. Any attempt to alter a past record—such as changing a vote—would break these links, instantly exposing fraud.

Storing data on a blockchain also implies storing code, meaning it can run programs.

Blockchains can execute self-enforcing code (smart contracts), which automatically enforces rules—for instance, releasing funds only if a proposal gains enough votes—eliminating the need for a central authority.

That offers a new way to make collective decisions—like votes or resource allocations.

By anchoring governance processes in a secure, decentralized ledger, we can move closer to the democratic ideals of transparency, fairness, and accountability.

The Future

Combining the four building blocks—direct democracy, software, open source, and blockchain—the world I envision is as follows:

Most of our institutions’ decision-making processes can be digitized and implemented as software running on a blockchain. This includes tax codes, criminal laws, policies for resource allocation, rules for promotions, and, most importantly, voting.

People then can propose direct changes to the code that run our government, similar to how software engineers contribute to open source projects today. Then, if the suggested change was approved by the community (via blockchain voting mechanism), it will enter into effect immediately.

The “software” is the underlying “physical laws” that actually perform the calculation on the blockchain - but most of the people will be using AI to both read suggested changes, and to write/modify the code that run our society.

—

Software is uniquely suited for the task because it is inherently iterative, enabling rapid adaptation to societal needs, and collaborative, allowing diverse contributors to build, review, and improve systems transparently.

Its efficiency surpasses traditional legal frameworks, with algorithms that process decisions faster and with fewer errors.

Software is verifiable. Every action or rule can be programmatically and mathematically proven to have occurred or not—ensuring precision and accountability.

—

For this system to function securely and fairly, the software must run on a blockchain. Blockchains ensure decentralization, preventing any single authority from controlling or altering the system.

They make tampering the data and the system nearly impossible, as every change is recorded across a distributed network and cryptographically secured.

—

As mentioned, AI systems are critical in making this vision practical. They lower the barriers to creating and managing software by enabling non-technical individuals to interact with and contribute to governance systems.

AI can translate complex programs and code into easy-to-read text, and vice versa. Making the direct democracy, through software contributions, accessible to everyone.

Conclusion:

The future of democracy is direct, exercised through open-source software running on a blockchain. By leveraging AI to simplify complex issues, this system can promote faster and more transparent decision-making. It directly addresses many shortcomings of our current indirect democracy—such as slow bureaucratic processes, corruption, and lack of public engagement—while restoring power to the people.

As software continues to ‘eat the world,’ it’s time for it to reshape the most important domain of all: how we govern ourselves.

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything.

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

This feeling almost has a smell and a taste in my mind as I think about my past. I remember being alone in my room, doing my own thing, exploring my thoughts. How I was not understood by others, from my actions to my thoughts. From having different wants and needs compared to my peers, to the way I expressed myself.

I was loved, I was hugged, and I was around great people and family. I felt happiness and I laughed. I felt closeness and warmth. And yet, my existence in this world was always accompanied by a slight feeling of loneliness.

And I appreciate this feeling, as it made me human. Loneliness brought me my closest friends and most meaningful experiences. It introduced me to the depths of knowledge in the world, and gifted me with the curiosity to explore it. It drove and drives me to build, to chase, to try. It made me seek love, and to try again and again, regardless of the pain.

Loneliness might not be the right word to describe what I am trying to convey, nor is solitude sufficient. But what I argue is that this “thing” is at the core of our free will, and the base of our civilization. And I will try to elaborate.

—

Loneliness is the essence of the human experience—it is what made our civilization.

You strive to construct a world around yourself— you seek connections and belonging through relationships, family, and friendships. You pursue love, spirituality, curiosity, and passions from the arts to science. You read, create art, write poems, compose music, and tell stories.

And our world as a whole, built by many humans, is not different. Institutions—old and new—are created to impose meaning on our lonely existence on this big, floating rock. God and religion, countries, nations, and governments.

When you go outside, or even explore your own mind—all you see is an endless effort to fight our lonely existence, both as individuals and as a species.

"It is not good for man to be alone" (Genesis 2:18)
...Thus, he builds a world around him.

Just because I deeply know loneliness, I am using all my energy to escape it.

—

Adam

Cameras recording every move, data aggregated at every click, social media monitored, bots and algorithms subtly steering our thoughts, shaping our so-called “free” will. Psyops at every corner, fighters criticized, canceled.

At the dawn of the Information Age, visionaries saw cyberspace as the arena where we would finally transcend the physical limitations that had long shackled us. It was supposed to be our sanctuary, our escape, our new frontier.

But as the 2000s unfolded, it became clear that the powers weren’t willing to surrender without a fight. They built efficient cyber teams, recruiting our best hackers. They installed cameras for our safety and monitored our phones for our own good. They sold our data and made a fortune. And with that fortune, they expanded their control, tightening their grip on the very freedoms we once had.

—

But, as in every good old American movie, we had heroes. Along the way, a few pioneers—true visionaries—saw what was happening and fought back. Thanks to them, we’re in a much better place than we could have been.

From the US Founding Fathers, who insisted on privacy in the Bill of Rights, to Whitfield Diffie and Martin Hellman, who pioneered public-key cryptography despite resistance from governments; from Tim Berners-Lee, who invented the World Wide Web and kept it open, to Phil Zimmermann, who faced government scrutiny when he created PGP encryption to protect our emails; from Linus Torvalds, who gave us Linux, to Satoshi Nakamoto, who introduced blockchain and Bitcoin, challenging centralized power.

“Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety.” — Benjamin Franklin

—

This is our turn now—what can we build?

Counter-Surveillance: Privacy is essential for true freedom. Constant monitoring and tracking can erode our personal autonomy. We must actively address these concerns to protect our right to privacy and ensure that surveillance does not compromise our freedoms.

Free Speech / Free Will / Psyops: True freedom and free will are at risk when we're manipulated, whether by targeted agendas or dopamine-driven addiction. Social media’s black box algorithms, which obscure how they influence what we see, can subtly shape our thoughts. We need to counter that trend, and reclaim our minds..

Education: The people—our fellow citizens—don’t care enough. They don’t understand the stakes. They don’t realize their Windows can be remotely controlled, their thoughts influenced without their consent. We must educate, put the information and the tools out there. We need to empower people to take action.

Thank you for reading.
Twitter: https://twitter.com/adamcohenhillel

Another example of such a monopoly, which is far less thought of, is the Monopoly on Identification - the monopoly on issuing identity documents, such as passports, national ID cards, driving licenses, etc. Identification documents are needed for buying alcohol in the shop next door, flying overseas, buying a house, getting married, opening a bank account, collecting taxes, enforcing criminal and civil laws, and countless other crucial activities. In fact, this monopoly is so fundamental to modern-day life that it can be used as an enforcement tool (as it sometimes does) and to undermine basic human rights (freedom of movement).

However, this state monopoly might be disrupted in the future, led by the challenges AI brings upon us. Identifying humans in the digital realm is becoming extremely difficult, resulting in the need for new and highly technical solutions and the increasing leaning of governments on tech companies to help as Identification Suppliers.

The implications of this shift (governments relying on tech companies for identification) are worth thinking about, and hence, I write about it. I don’t have solutions yet, but I am actively thinking about this.

In this essay:

• We will start by understanding why is Identification so important

• And why highly technical solutions are becoming an extremely crucial need,

• Then, we will see how the tech giants are positioned better to provide such solutions, examining the existing players and the new industry of “Proof of Personhood,”

• We will talk about the shift period and what it might look like,

• And lastly, we will see why this can be problematic

The modern state, Identification and AI

Identification is essential for the functioning of a modern state. Without it, we cannot efficiently collect taxes, raise an army, distribute welfare services, invest in infrastructure, resolve disputes at scale, or operate any system that, well, requires the use of identities.

In the past 20 years, our society and all these activities have increasingly transitioned to an online and digital landscape. This shift, combined with recent advancements in AI capabilities, presents a new challenge: advanced AI systems that can be indistinguishable from humans.

Leaders and tech giants worldwide are raising concerns about AI and bots flooding the internet. These concerns highlight the potential for the internet to become overwhelming for humans, making it difficult to identify whether they are interacting with a real person or not. This situation poses significant challenges for maintaining trust and security in our digital interactions.

New industry: “Proof of Personhood”

A new technological space called “Proof of Personhood” aims to address the growing problem of sophisticated AI bots by creating a digital infrastructure to prove human identity online.

One of the more known players in this space is Worldcoin, led by Alex Blania and Sam Altman. Their system uses biometric technology (iris scans via a device called the Orb), to create unique digital identities on a blockchain, known as World IDs (It is a cryptocurrency, where each human/iris get one World ID).

“More than three years ago we founded Worldcoin with the ambition of creating a new identity and financial network owned by everyone; the rollout begins today. If successful, we believe Worldcoin could drastically increase economic opportunity, scale a reliable solution for distinguishing humans from AI online while preserving privacy, enable global democratic processes, and eventually show a potential path to AI-funded UBI.” — Worldcoin

Apple:

Though not yet officially in this space, Apple is probably positioned in the most ideal place for winning this transition, and enable the world with digital proof-of-personhood, for a few reasons:

1. They already have the biometric technology to identify humans digitally (Face ID, Touch ID, veins structure via Apple Watch)

2. This technology is already distributed across the world (in 2023, Apple surpassed 2 billion active devices!)

3. Apple has a well-established commercial relationship with the US government, providing iPhone to US officials.

   

4. Apple, for a long time, is trying to position itself as the privacy company.

   

So, What’s the issue?

The main issue with this growing trend, of commercial companies going after human identification, is that - if they will own this industry, and will be the go-to providers/suppliers of “Proof of personhood,” it will give them an unprecedented power against nation states, and its citizens.

Worldcoin is directly aiming to be part of the democratic processes (“enable global democratic processes, and eventually show a potential path to AI-funded UBI”) - now, if that will be the case, if they will define who’s a human and have rights, and who’s not - then what?

By giving up that crucial component of their monopoly, nation states might decline in influence and power, to a non-democratic entities, the commercial companies.

—

Who vouch your personhood?

Right now - it is the state. They give you an ID.

But, with the speed of AI development, we must ask the question - can countries actually be efficient enough to roll out the needed technological solution for the AI problem? Most of the passports in the world today are still not biometric, while Apple, right now, can just roll out a new software update, issuing IDs to their iPhone users. These IDs will be more reliable than most passports currently being used to cross borders all around the world.

—

(I am not saying the state should have this power either - but definitely not a for-profit entity.)

Not Unprecedented

When I mention to people my concerns of how a commercial entity can overthrown a state, they see it as a too far-fetched, like “there is no chance, the government can do whatever they want, they can take over Apple if they want to.”

To them, I say - go read about the East India Company (EIC).

EIC literally overthrown the Mughal Empire in India in 1757, at the battle of Plassey - a moment where a commercial corporation first acquired real and tangible political power. At one point, EIC held a private army of 260,000 soldiers, twice the size of the standing British army (!!!).

In today’s modern world, power is not necessarily the number of troops you have. The power that is being given to corporations should be thoroughly thought-of.

What can we do?

As I said at the beginning - I am not sure.

I do think that first step is to make sure these technologies are being developed as Open Source (which Worldcoin seem to be committed to). This will create a somewhat democratic nature into their system, and will allow us, humans, to take part in the decision making of the technology, or at least branch-out.

Another thing we need to think about is this new space of technology for protecting your biometric data. You will need to protect it after it was collected into a digital form, and to protect it from thief in-real-life.

Lastly, we need to find a better way to digitalize and adjust the three-centuries-old democratic system to the age of AI.

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

In short, Mustafa claims that the upcoming technologies of the next few years (mainly in the space of Artificial Intelligence and Biotech) will be so fundamentally disrupting that we must change the way we allow and conduct innovation, scientific discoveries, and knowledge sharing - all in order to contain the danger these technologies will bring. The book is an ongoing case for why we should do that, how this wave is different from previous technological changes, and eventually finish with what we must do to be safe and for civilization to continue.

In this essay, I hope to debunk the holes in his ideas and show why they’re problematic; we will talk about:

1. Why nation-states are not our strong pillars

2. How the well-funded companies use regulation as a moat

3. The upcoming unipolar AGI scenario

4. Unwanted side-effects of regulation on individuals

5. The issue with AI-safety research on out-of-reach black box

6. Why closed-source security is not superior to open-source

7. The biased progress in AI alignment

Then, I will suggest an alternative approach for moving forward in these changing times. Let’s get started!

Issues with Mustafa’s Arguments

Wrong Starting Point: Why Nation-States Are Not Our Strong Pillars

Before we dive into the more technical holes of Mustafa’s ideas, I want to briefly refer to the most fundamental building block of his solution argument: the nation-state.

Mustafa uses the nation-state as the de facto tool for world cooperation and world governance to fight and save us from AI that goes rogue from civilization’s best interests.

However, I disagree with the underlying assumptions here that nation-states have the public’s best interests in mind. In today’s Western world, democracies are more polarized than ever. Trust in institutions and leaders is at a low point, while corruption is at a high point worldwide.

And the list goes on and on. But unlike Mustafa’s claims that “as an equal partner in the creation of the coming wave, governments stand a better chance of steering it towards the overall public interest,” - it’s enough to consume the media in the last few years to realize how corrupted the world is. From Biden’s son, Hunter, and his legal troubles to the broader phenomena with the Twitter Files - governments and public officials are not aligned with the public’s best interest as we wished they were.

Suppose this technological revolution is as big as Mustafa suggests- a breaking point from which there is no going back, a powerful tool that puts everything we know in danger. In that case, governments are not the ones to handle it, at least not in their current structure; they just don’t have the moral mandate to make AI-related crucial decisions on our behalf. Accumulating such power in the hands of the few won’t end well.

How Well-Funded Companies Use Regulation as a Moat

Throughout the book, Mustafa tells us stories about how previous regulations worked amazingly. From the nonproliferation of Nuclear Weapons, the FAA (Federal Aviation Administration), the FDA (Food and Drug Administration), and a few more examples. He uses these examples to make his later suggestion land better: a license-based regulation for AI companies:

“Today, anyone can build AI. Anyone can set up a lab. We should instead move to a more licensed environment. […] so you shouldn’t simply be able to release a state-of-the-art AI,” Mustafa Suleyman

Now, there are two terms here I want you to understand:

1. “moat,” which, in the context of businesses and companies, means a company’s ability to maintain competitive advantages over its competitors to protect its long-term profits and market share from competing firms.

2. “regulatory capture” - a situation in which a government agency that was meant to regulate an industry becomes too influenced by that industry, undermining its ability to protect the public interest effectively and ending up serving the interests of the industry itself (and the big players in that industry in particular).

Regulatory capture is already being used as a moat by many industries. Companies are actively trying to leverage and manipulate regulations and policymakers. Special interests are shaping policy for their own benefit, stifling competition, raising prices, and inhibiting innovation.

The following talk by Bill Gurley is a great explanation of how companies throughout the years have done exactly that. Bill explains this better than I ever can, so I highly recommend you watch it. But, if we fetch just one example from the video, it will be The Telecommunications Act of 1996. This was a US legislation that attempted to bring more competition to the telephone market for both local and long-distance services. However, the result was that over 20 states essentially outlawed cities from providing free public WiFi due to telecom giants like Verizon and Comcast, who lobbied state legislatures to pass bills banning municipal broadband, protecting their own commercial interests rather than serving citizens.

With that knowledge in mind, I’d like to note how well-connected Mustafa is in both the UK and the US governments. Just recently, he met with US President Joe Biden to discuss the dangers of the AI revolution; he has connections among the leading AI organizations like DeepMind by Google (which he founded and was acquired by Google) and other influential individuals (like Yuval Noah Harari, Eric Schmidt, and others).

I am not after attacking Mustafa specifically, but whoever thinks there are no special interests involved in these discussions is delusional and naive.

The Upcoming Unipolar AGI (Artificial General Intelligence) Scenario

Another issue with regulation is its outcome. Whether you agree that an AGI (Artificial General Intelligence) is on the horizon or not, what matters for the sake of this argument is that the people who are leading the big AI companies do believe (or claim to believe) it is coming. From the book’s author, Mustafa, to OpenAI’s Sam Altman, Elon Musk, and others.

Now, there is a common discussion among AI researchers about what happens when we achieve AGI - is it better to have one AGI (known as the Unipolar AGI scenario) - so we can more centrally control it, or are we better off having a few (Multipolar AGI scenario) - so that the power will be balanced and distributed?

Suppose you are like me, believing in democracy and the idea of Separation of Power - then, obviously, the multipolar scenario sounds better. But, there are varied arguments for either approach, with some philosophical research and discussions being done in that space, with discussions on Twitter, too. Overall, though, the leaders of the AI community believe that the multipolar scenario is a favorable outcome for humanity. From OpenAI’s CEO Sam Altam, who said that “multiple AGIs in the world I think is better than one,” to Elon Musk, who said in an interview that “you don’t want to have a unipolar world where one company dominates in AI.”

So we have people who are saying the following three arguments simultaneously:

1. AGI is coming

2. Having one is bad; having a few separate ones will be better

3. Let’s regulate our industry and create an entrance barrier

The issue is that we won’t get a multipolar AGI scenario with a well-coordinated, unified world government that regulates and places high entrance barriers for AI research and development. They cannot have it both ways. They cannot morally claim that we must have a multipolar AGI scenario while also restricting actors from participating in this space. It just doesn’t work. And as Yann LeCun, Chief AI Scientist at Meta (Facebook), says, “[It will] *inevitably* result in what you and I would identify as a catastrophe: a small number of companies will control AI.”

Unwanted Side-effects of Regulation for Individuals

Last point regarding the regulation issue: President Biden's recent Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence shows that they plan to enforce AI safety by limiting the amount of compute power (hardware, processors, computers, essentially) you can own and the amount of operations you can perform on it ( FLOPS - floating-point operations per second).

Currently, the suggested limit is very high, targeting large companies and their enormous datacenters. However, today’s compute power of the best supercomputers is tomorrow’s consumers’ level of compute power. The compute power that was exclusively delivered by supercomputers will be accessible to startups in garages and individuals in the next few years - as Moore's law has consistently shown us over the last 50 years.

Rolling back regulation after it passed is not an easy task, and it sets a precedent for further regulations and tightening of the restrictions (which, in this case, basically means limiting math).

Additionally, regulations can be used maliciously by the government itself - like what happened with the Patriot Act in the US. In the name of fighting terrorism after 9/11, the Patriot Act allowed the government and law enforcement to perform mass surveillance on citizens - which they actually did. The AI act might enable governments to misuse their power in a similar way, and much worse: limiting individuals’ privacy (to track who owns what hardware), autonomy, and liberty (what you do with your computing power).

Why Closed-Source Security is Not Superior to Open-Source

To show how previous technologies were maliciously used to cause harm, Mustafa gave an example from the realm of cybersecurity, a space I’m happy to have some knowledge of!

He talked about how cyber attacks (in particular, WannaCry, a ransomware attack that disabled critical infra in the UK a few years ago) - showed us that “core institutions of modern life are vulnerable” (I’m quoting Mustafa here, yeah?) as “such attacks demonstrate that there are those who would use cutting-edge technologies to degrade and disable key state functions” and that “lone individual and a private company (Microsoft),” had to patch it themselves.

Essentially, he is referring to the fact that the attackers used a loophole (known in the industry as a “vulnerability”) in Microsoft’s software Windows (the operating system) to block access to the computers of the UK NHS (National Health Service). He said: “If everyone has access to more capability, that clearly also includes those who wish to cause harm […] Democratizing access means democratizing risk.” - so far so good, isn’t it? Well, what he forgot to note is that in the case of Operating Systems, Linux, an open-source alternative to Microsoft’s Windows, is much more secure and often patches way quicker than Windows, and developing such attacks for Linux is much more complex.

The reason why Linux is more secure is because everyone has access to the underlying code. Therefore, everyone can find and fix security risks (which are then all publicly available). Because of the nature of that software paradigm, there are plenty more people in the open-source community who are working on finding and fixing these bugs than you have working in Microsoft (across all of the company!). They often do it for free (for the sake of the community, humanity, or just curiosity) - and also through paid bounties (plenty of companies have the incentives to find and fix open-source vulnerabilities in Linux, as it is used in so many places. In fact, it is usually one of the most well-paying bounties, as it is also one of the toughest to find issues in due to its hardcore security)

The argument is simple: creating models openly leads to a more resilient solution that is less prone to harm than closed development, AI safety concerns, AI tools, etc. - are not an exception to that rule.

(Additionally, similar to what we previously talked about regarding special interests when it comes to big corporations and the government - it also works the other way around. Microsoft is working so closely with the US government—one of its largest enterprise customers (if not the largest). It wouldn’t be a surprise if the US gov approached Microsoft (or even agreed on a formal procedure) to “not” patch publicly undiscovered security risks for some time (these are known as 0-day) so the government can use it in the name of “national security.” Microsoft has great incentives to follow and not to fix some issues - working against the public best interest, as the US government is the biggest buyer of zero-day vulnerabilities, report claims.)

The Issue with AI-Safety Research on Out-of-Reach Black Box

Another point that Mustafa is making is the need for AI safety research. He suggests that the government should make companies dedicate 20% of their R&D budget to safety research.

I am not ignorant of the potential great risks that AI can pose, from the social instability associated with mass labor automation to the age of misinformation, the lack of establishing truth online, and even the doomsday scenarios of paperclip maximizers. They all have a place in the discussions.

To contain these risks, or at least roll them out from feasibility - we need to understand these models as best as possible. We need to understand how they work, what neural structure is created in them during training, what biases and pitfalls are fused to their weights, etc.

Therefore - I agree with Mustafa that there is room for AI safety research and a place for the government to incentivize that.

However, I can’t see how we can have effective AI safety research without access to the underlying systems. Mustafa’s call for safety is within the organization, but the best safety research I’ve seen so far came from the community, investigating an Open-source Large Language Model called Llama-2!

There is a reason open science and knowledge sharing was and is so fundamental to the flourishing of humankind - it’s a way to distribute difficult and complex tasks most efficiently. No matter how much money you pour into AI safety research (20% like Mustafa suggested and OpenAI committed to) - you cannot get better results than the entire AI community worldwide. You just can’t.

The Biased Progress in Closed-Source AI Alignment

Lastly, I want to add context to one of the great achievements in AI alignment that Mustafa mentioned in the book - the “RLHF” technique (Reinforcement Learning from Human Feedback). This technique trains AI models by having humans give feedback on the model’s outputs, which the model then uses to improve.

However useful it is (it was one of the main improvements in creating ChatGPT), the humans providing feedback are inherently biased. Right now, they are likely Western English speakers - in fact, we can already see that ChatGPT leans liberal. Diverse global perspectives are not incorporated into this process. For actual human alignment, RLHF needs more representative human feedback, which requires giving access to the dataset and the training process to the entire world.

So, what am I suggesting?

It is much easier to debunk someone’s ideas than to come up with your own. And I must say I am not entirely opposed to Mustafa’s ideas and arguments, as already mentioned. Some of them are relevant, and I think he generally has good intentions.

But, his proposed solutions of increased government regulation and closed-source development are fundamentally flawed. As discussed, our current governmental institutions are plagued by corruption, partisanship, and regulatory capture by special interests. Handing over even more power to regulate technology is unlikely to serve the public good. Additionally, closed-source development limits transparency stifles innovation, and consolidates control in the hands of a few powerful companies. Mustafa’s arguments for AI safety research and multipolar AGI also contradict his calls for regulation and closed systems.

Rather, we should look to decentralize control and democratize access to powerful technologies through open-source development, decentralized governance models, and a diverse, multipolar landscape of perspectives and values.

1. Openness: The world is undoubtedly a complex place. Trying to capture and steer it with policymakers and high-profile innovators alone is doomed to fail. It is the same as the free markets, which work because they are open and free, whereas central planning failed, as it is too complex. AI risk containment is similar; it can only work in the open, where humanity as a whole works on that together.
   
   Openness, therefore, stands as a cornerstone of our strategy. I encourage any initiative that supports, cultivates, and encourages openness in the space of AI research, safety, and alignment - government or otherwise. Governments can put more pressure on companies to open source their datasets, architectures, and weights so that we, the community, can investigate them, find the different pitfalls, stretch their limitations, patch the issues, develop safe mechanisms, and build tools that help us monitor anomalies in the models; I encourage the government to do that in the open, too. As a community, we need to work in the open, share our findings and insights, help each other develop tools and ways of evaluating dangerous behavior, and monitor progress.

2. Open Safety Research: As mentioned previously in the essay, the best way to face the risks is to understand these AI models as much as possible. We need to know how they work internally, what different neural structures are responsible for what capability, etc. As a community, the industry should put more value and attention into AI safety and achievements. OpenAI’s initiative to create bounties for bugs is a good start, but we should do more: bounties, rewards, grants, etc. We should fund and encourage open investigation of AI models. This is something that the government can and should incentivize, too.

3. Tools: When it comes to the more urgent issues with AI: misinformation, manipulation, deep fake, etc. - I think the only option we have to fight it is *more technology*. To have a society where “deep fake video” is not an issue, we need to create tools that help individuals detect them rather than trying to stop people from creating them. This can be achieved, again, by government grants and such - but I also believe it will naturally be created even without govs because the need will be there, and the free markets will fund it! (Unless, of course, we will be regulated from creating the solution to the problem)

4. Red Teams: Here, I am very much aligned with Mustafa. I think government-funded red teams will be useful (a red team is a group that pretends to be an “enemy” / a malicious actor and tries to “break” the AI model to find its flaws). However, their findings must be available online for everyone to learn from.

5. Some Place for Regulation: Another interesting point I saw recently on Twitter was by Clem, HugginFace’s CEO, where he suggested that governments should regulate based on use cases and not based on the size of the model and the amount of compute needed (as currently suggested by the US):

   

   This can allow the development of good AI that pushes humanity forward while also avoiding potential bad use cases. That being said, transparency and open discussion on what these industries are are important.

And Lastly - A few Words About the e/acc & Techno-Optimism Counter-Movement

There is a counter-movement to the established “calling for regulation” AI leaders called e/acc (Effective Accelerationism) and/or Techno-Optimism. In short, they’re saying that we must keep pushing (and accelerate) scientific research and technological innovation forward as much as possible and that we should not hold it by creating a high-regulated environment in order to keep civilization flourishing. As you can tell by my essay, this is also my view (more or less).

However, I disagree with something about this movement, and I felt like I should share my perspective here, hoping some of these people will read it.

While claiming to advocate for a distributed power and acceleration (as they recently have, calling for an alliance with the crypto community), which is indeed a crucial aspect for the positive future of humanity and democracy, they are also very much following the problematic aspect of the existing establishment by having a leader-first paradigm. Founders-centric movement like that tends to concentrate powers, hierarchy, and longer-term also unproductive bureaucracy. And it is funny because Marc Andreessen, the author of the Techno-Optimism manifesto, is saying clearly that “Our enemy is bureaucracy,” - yet he glorifies the “Patron Saints of Techno-Optimism.” - the irony is a bit sad, as status games, hierarchy (saints), etc., are what create an unproductive bureaucracy!

So, for the e/accs people who might read this: While pushing for the advancement of civilization, beware of creating the same thing you’re trying to replace, if not worse. The leadership and the powers that drive the movement forward should also be decentralized.

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything that was mentioned here!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

Today I will walk you through my experiments with different Agents I created (coding, entertainment) and share my view on the future of this technology.

Playing around with the original BabyAGI version, I quickly realised its great potential. The reason is that recently I noticed that the way I use ChatGPT (the web interface) is not the same way I used to use the “old” internet (querying questions/problems and getting answers), but rather - I use ChatGPT as cognitive offload. I provide it with tasks (mostly coding) that I know I can solve myself, but I rather spend my time doing other things. When I saw BabyAGI - I saw how it could enhance my ChatGPT flow by 10x and more!

So I took a step ahead and started implementing different agents:

My Experiments

Experiment #1: TDD (Test Driven Development) Coding Agent (Twitter thread).

TLDR: Coding Agent that follows the Test Driven Development (TDD) methodology! You write the tests - and the agent runs in a loop until it creates the feature properly!

To integrate autonomous agents into the workforce, I realised, we’d need to be able to measure their progress, recalculate their direction, and flag when they finished/need help. Essentially, having a framework for it to work within, like setting expectations with a new junior dev. This was when TDD came to mind! In the case of coding agents, it can act as the agent’s framework.

💡 To those of my readers unfamiliar with the TDD term - it is a software development practice that focuses on creating test cases before developing the actual code.

So what do we see here?

1. We’ve got an incomplete application code (in this case, a simple FastAPI app) and two tests. One of the tests fails (as the application is incomplete).

2. We then run the agent - it can’t read the tests but just run them and get the output.

3. On the first try, it created the missing "echo" endpoint with a POST method - which failed, as the test defined it as a GET.

4. It then took the failures and converted the code to use GET instead - which failed again but with a different error (we tried to make it fail on purpose, so in the test, we defined the expected key name of the return value as "message1" and not "message")

5. It, therefore, took the failures (again) and adjusted the code - this time making the return value "message1" - which worked, and the tests passed.

The idea here was to show how TDD is super useful to steer it in a specific, measurable direction - while keeping it somewhat autonomous.
As a software engineer, to leverage agents in my workflow, I have to have a way to steer the agent to my desired end goal, and whatever it does to get there - is up to it.

Have we finally found a good reason to follow the TDD practice? ;)

Experiment #2: Dynamic Chatbot that creates its own inner world (Twitter thread).

In a completely different direction, I wanted to explore the potential entertainment side of autonomous agents. And yes, I recently re-watched Westworld, which affected my thinking in this experiment :)

While watching the show, now with the knowledge of LLMs, I realised that what attracts people to come to the park (take all the crazy inhuman things aside) are the dynamic loops the “hosts”/”agents” live in. Unlike today’s interactive AI products, Westworld ones have their own lives going on, even when visitors aren't around. It's not just about them waiting for people to interact with them.

Can I use LLMs to do something similar?

So I created “Dreamer” - an autonomous chatbot. You provide it with a given personality, and it then runs in a simulated environment and can do whatever: exploring, chasing interests, widening its knowledge, playing games. You can chat with it whenever, but it doesn't sit and wait for your input! It does its own thing :) dynamic chatbot with evolving inner world!

So what do we see here?

1. The agent is given the personality. “Like to play board games, explore new ideas and read science fiction.”

2. It then starts by exploring the internet for good science fiction books to read, which comes up with a few results (i.e. Dune by Frank Herbert).

3. It chooses to save this to its memory for later.

4. It does the same for board games^

5. The user then asks it, “Hey what you doing? :)”

6. It tells the user about what it was up to

7. It continues to create a new type of boarding game

8. etc

In other runs, it also simulated a Catan game, reading a book and writing its opinion, etc. This Agent is a bit simpler than the coding one, as the accuracy/precision of its output is not as critical (see the “Takeaways” section below for more details), but this is also why it is a super interesting use case - it can get wilder!

Check this out (link)→

Experiment #3: LLMitlessAPI - one single endpoint is all you need! (Twitter Thread)

This experiment, for some reason, was the most difficult to convey to the readers - so I will try my best here.

Most (arguably all) backend API services are a combination of these 3 operations:

1. Store

2. Fetch

3. Execute

You can combine them in infinite combinations and create any API out there. So what if I place an agent that knows how to do these 3 things behind one API endpoint? We can then ask it to act as if it was a different endpoint every time!

Define the service you want the agent to act as + the data to act upon. That is it. limitless API! (Or should I say, LLMitlessAPI? ;)

So what do we see here?

1. Nothing in the backend/API defines a “chat” service;

2. The client sends a message to the API asking it to act as a chat API, + the message send

3. The agent then decides to store it for later, for whoever will ask for it

4. Another client asks the API to act as a Chat API too, but asking for new messages

5. The agent fetches the latest message from its memory, sent by the other client, and sends it back

6. BAM - a working chat (with many bugs and very slow) - without creating the chat functionality!

I don’t see much “production” value facing customers for this use case. Still, it can help software engineering teams to iterate faster by creating a working PoC of products in a fraction of the time, getting feedback quicker and building what works, and throwing out what isn’t. Founders can also use it to create a demo and raise money, etc.

Takeaways - Agents in Production?

An autonomous agent is a very interesting concept worth more experiments. Still, I think they are not yet production-ready. For most of the use cases, we’d need to develop a strong framework for the agents to work within before we can gain value from them (a simple proof-of-concept of a framework was the TDD agent, which has the potential to be further improved for production coding agents).

The diagram below describes this well, but to deploy an agent to production use cases, we either need a good framework to measure the agent progress and output, or the use case doesn’t requires too much precision.

Some use cases are more “frameworkable” by nature, like coding, where you can “verify” correctness more systematically by writing tests or running the code and seeing what it does. Whereas use cases like therapy are more open-ended, and the results are not [easily] measurable (even in human-to-human interaction, it is difficult to measure the progress/impact of a session, not to mention long-term effects).

I believe gaming (and entertainment in general) will be the first production-use-cases in the upcoming months (due to low “precision-critical”, even if difficult to verify), followed by coding agents (due to high “ease-of-verification”).

Exciting times!

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

In the conclusion of that post, I noted a few issues with using such a feature in production environments. In particular, sharing highly-sensitive data with a 3rd party API is a big security & privacy concern (see more about how Employees Are Feeding Sensitive Biz Data to ChatGPT, Raising Security Fears)

Since then, OpenAI has changed their API data usage policy, announcing that they will not use data submitted by customers via their API to train or improve their models. Even still, they retain any data sent through the API (for abuse and misuse monitoring purposes) for a maximum of 30 days. This, therefore, is still a big concern for any app or feature built using their API (See, for example, how ChatGPT bug leaked users' conversation histories).

Introducing Masked-AI

Masked-AI is a Python SDK and CLI wrappers that enable the usage of public LLM APIs such as OpenAI/GPT4 more securely. It does this by

1. Replacing (“masking”) sensitive data (e-mail addresses, persons names, credit cards numbers, etc) with a placeholder in its place.

2. Storing a lookup table for the placeholders  -> data, to be used later to reconstruct the response

3. Sending the request to the API, with the masked, “safe” data

4. Replacing/reconstructing the sensitive data back into the output

The result is that you get the same output from the API, without having to send the sensitive data. Simple but powerful, here is a diagram:

Here is a simple example of how the CLI tool works with the basic ‘echo’ command:

You can deploy Masked-AI straight from pip (“pip3 install masked-ai”) or from the GitHub repo: https://github.com/cado-security/masked-ai. It can be used as both a python library or over the CLI (binary available in the GitHub repo release).

Using OpenAI Completion API, with Masked-AI CLI tool:

./masker --debug --prompt "Hello, my name is Adam, say my name" curl https://api.openai.com/v1/completions  -H  "Content-Type: application/json" -H "Authorization: Bearer <OPENAI_API_KEY>"  -d  '{"model": "text-davinci-003", "prompt": "{prompt_placeholder}"}'

Notes:

• Don’t forget to change `<OPENAI_API_KEY>` to your own OpenAI key

• The string `{prompt_placeholder}` in the curl command is where your safe, masked `--prompt` will go.

And the output:

So, what is happening here?

1. If we look at the output, the prompt that the user wanted to send to the API was “Hello, my name is Adam, say my name”, but what is actually being sent (marked with blue) is `Hello, my name is <NamesMask_1>, say my name`, Masked-AI replaced the name “Adam” with a placeholder.

2. Then if we look at the raw return value from the cURL command (the important part is marked in red), we can see that OpenAI returned the following completion: `Hello, <NamesMask_1>!"` 

3. Lastly, the reconstruction stage (marked purple), where Masked-AI takes the output, and replaces the placeholders back with the real data (using the lookup table), which in this case, `Hello, Adam!`

That’s a simple example showing how we can still use LLMs, and leverage their great power, without sending out sensitive information. Here is the same example, but with Python:

import os
import openai
from masked_ai import Masker

# Load your API key from an environment variable or secret management service
openai.api_key = os.getenv("OPENAI_API_KEY")
data = "My name is Adam and my IP address is 8.8.8.8. Now, write a one line poem:"
masker = Masker(data)
print('Masked: ', masker.masked_data)
response = openai.Completion.create(
    model="text-davinci-003",
    prompt=masker.masked_data,
    temperature=0,
    max_tokens=1000,
)
generated_text = response.choices[0].text
print('Raw response: ', response)
unmasked = masker.unmask_data(generated_text)
print('Result:', unmasked)

How and What Masked-AI Masks:

Masked-AI currently masks: persons names, credit card numbers, email addresses, phone numbers, web links and IP addresses, using either Regex or the NLTK library. We also made it very easy to contribute to the code and add more masks by adding just 4 lines of code, check out How to Contribute.

Other approaches

In an ideal world, we’d be able to run these models ourselves on our own compute, expensive as it might be. This might come available in the future, thanks to Microsoft's (Azure) partnership with OpenAI. But until this happens, there are a few open-source alternative models, such as Meta’s Llama, which with certain configurations, can even run on a mobile device. However, this is not as production-ready as OpenAI APIs, the community is not as large, and more applications and repositories have been built for GPT models.

Conclusion

Large Language Models are going to revolutionize many fields, and this is exciting. It is here to enhance our productivity, and in the case of incident response, it allows security teams to focus on the more important (and fun) parts of investigation and response. However, as always, there are potential issues that may arise, such as data privacy, that should be taken very seriously.

You can get masked-ai at https://github.com/cado-security/masked-ai (Open sourced python package, and the compiled CLI version). And if you’re interested in checking out Cado Secuirty's latest integration with ChatGPT to further expedite incident response in the cloud, you can deploy the free trial here.

This blog post was originally published on the Cado Security blog here:
https://www.cadosecurity.com/introducing-masked-ai-an-open-source-library-that-enables-the-usage-of-llm-apis-more-securely/

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

TL;DR

1. AI will not replace but enhance software developers to be more productive and to create products that weren’t possible in the past.

2. Boost in productivity occurred many times in the past - at higher rates than we currently see with AI tools—E.g. with high-level programming languages, the shift to the cloud, open source culture, etc.

3. The barrier to writing code will be reduced, and more people will enter the arena.

4. All of the above is a positive trend as it will boost the industry, not the other way around. It will lead to a new era of growth, products and opportunities for all.

5. To be part of it, you do need to be active, keep updating your skills, and embrace new tools.

Let’s GO:

It is crucial to begin by acknowledging that the recent boom of AI tools, such as Copilot by Github, Codex by OpenAI, and ChatGPT by OpenAI - have shown us very clearly, that they are capable of enhancing software engineers. Not the potential to, but the already-here usefulness. I'm using ChatGPT daily, and it almost entirely replaced my Stackoverflow-usage. I write code for around ~8 hours a day, including weekends (for work and personal projects), and I’m saving about ~5 hours a week thanks to it (based on an experiment I did).

However, the jump from where those tools are currently at, to the utopian/dystopian world of software engineers being unemployed is far-reaching.

For this argument, I’d refer to the much-beloved Pareto principle - or 80/20 Rule - which states that 80% of the deliverable from a project is achieved with 20% of the effort and that the remaining 20% will require 80% of the effort. In our scenario, this can be thought of in two ways:

1. The remaining “20%” for Copilot/ChatGPT to automate software engineers entirely will be the difficult 80% effort. Which means we still have a long way to go. This will be the reasoning, debugging, and the “connecting the dots” of different parts of the codes in a codebase (and less about the specific known algorithms, tests, etc).

2. Copilot automates the 80% “copy-paste” job of engineering, where we, the engineers, are left with closing up the remaining 20% - Yay! (More about it later)

Another way of looking at these tools is in the context of previous technological advances in our field. For example, Github argues that engineers using their Copilot are 55% more productive/fast in completing a specific set of tasks (note that this might not be fair research as the tasks might’ve been part of the model’s training dataset). This sounds like a crazy increase in productivity (and it is), but also not something we haven’t seen before. In the 1960s, with the introduction of High-Level programming languages, we saw more or less the same boost in productivity of engineers writing code in High-level languages compared to low-level languages.

A different concern I’ve heard is that this will make more people go to software, as it will be more accessible, making it much more difficult to find a job, make a living, etc. More people will indeed do software, as it will (and already is) reduce the barrier to becoming a software engineer. But it also opens up a new frontier of potential products, opportunities, and economic growth (As mentioned in my previous articles here and here). “We are at a *very* unique moment—on the cusp of a revolution that will usher in multiple trillion-dollar companies and industries”. Just like when the PC, the Internet, the mobile, and the cloud were introduced and were available for more people, it didn’t make people absolute, but it created so many new opportunities in the form of roles, companies, etc.

It is important to note that we won't be doing what we currently do in 5 years, which is okay. This is what this industry has been doing from day 1 - creating more and more layers of abstractions to enable us to focus on what matters - building stuff. For example, you don’t need to write your own authentication system nowadays. Thanks to companies like Google (Firebase), AWS (Cognito), etc., you can do that in 10 lines of code. Same with the cloud, you don’t need to care much for the scaling of your app, but AWS/Azure/GCP give you many different tools that enable you to scale stuff for millions of users effortlessly. In the past (~10 years ago), you needed to run your own server on your own hardware - you know how difficult it was? Thanks for not having to do that - we’ve experienced the boom of the 2010s with the mobile revolution. AI is another layer of abstraction that will enable you to focus on the business itself, building products faster and better.

Sam (OpenAI CEO) is a bit dystopian for the individual, but remember, he is the merchant trying to make his commodities worth more. The way we will write software in the future is somewhere in the middle. We will all have an AI assistant, some sort of ChatGPT, where it will be back-and-forth of English to code to English. We will be guiding it, and it will be guiding us back to the code. We will need to understand the code it generates and tweak it for the specific app. We will need to understand, at least conceptually, different algorithms and underlying computer science thinking (at a much higher level than today’s CS students, though) to be able to read and approve code (and mainly tests), and to “get under the bonnet” when needed. We will write more application code, connecting different interfaces (like using a 3rd party library or an API today) to form the required business logic. Understanding the code will be needed to do this reliably and at scale. You, as an engineer, will be able to do more stuff by supervising and guiding the tool and jumping in where needed. Think about yourself as a lead/senior engineer working with 5 junior engineers - this will be more like that than “give me an app that does X Y X” and that is it.

So, what should you do today?

All of the optimism above does not come just about, and you will need to be ready to be flexible, to learn, and to adjust. I am far from being an expert, but here are my recommendations on how to best face the changing world:

Start using these tools. I was surprised to hear how many engineers read about Copilot but have yet actually tried to leverage it. This will already boost your productivity and will enable you to do more. Not using those tools is like trying to implement your authentication system or run on your own server home at for production - you can, but why would you? Focus on the important stuff.

Be up-to-date with what is going on. That means reading articles (like this and many others). Follow and engage with relevant people on Twitter (from big names like the ones I embedded in this article to smaller ones like myself and many others). Being part of these discussions will help you form your perspective on the subject and to be in a better place in the changing world.

Learn & Build - Don’t be stuck with your current stack (whoop a rhyme!), keep learning new technologies that are released, learn about ML/AI/LLMs, build small projects and share them online, and be involved in the Open source community.

• For software developers without a background in machine learning, I will recommend watching Andrej's playlist here: https://www.youtube.com/playlist?list=PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ

• Reach out to me here or on Twitter and I’d be happy to share more resources and topics to learn!

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

Sources:

1. http://page.mi.fu-berlin.de/prechelt/Biblio//jccpprtTR.pdf

2. https://arxiv.org/pdf/1409.0252.pdf

3. https://en.wikipedia.org/wiki/High-level_programming_language#cite_note-kleith-2

Below we run through how the feature works, some interesting examples, and some caveats.

Automating Incident Response

For my readers who are not familiar with Cado, we are building a platform that empowers security teams with a smarter and faster way to respond to incidents in the cloud. To this end, we recently released Cado’s Automated Investigation feature. Automated Investigations help security teams seamlessly identify the root cause of an incident by automatically analyzing a robust dataset captured across cloud-provider logs, disk, memory and more:

Now, to further augment the investigation process, we’ve added an optional, experimental GPT-3 capability to Automated Investigations. This Automated Investigation first filters down millions of events and files from a system into a smaller set of key events that are relevant to the investigation. Then we leverage GPT-3 and semantic analysis to add the ability to ask interactive questions of captured evidence and receive a rapid response. 

An Example: Interacting with GPT-3 within the Cado Platform

As an example, we imported Cado’s Capture-the-Flag exercise, which recreates an incident detected by AWS GuardDuty. The attack impacts an AWS EC2 instance that accessed a known BitCoin mining address. The CTF data includes the AWS GuardDuty logs plus a disk image of the instance in question.

Now, let’s see what conclusions GPT-3 can help us draw from the imported CTF data. Below are a few screenshots showcasing how we interacted with GPT-3 within the Cado platform and its output. As you can see, GPT-3 helped us quickly understand how the system was compromised and other critical details:

While GPT-3 provides a good overview, we still have the ability to dive into the raw data collected by Cado to validate these conclusions:

We recommend that users do this as GPT-3 is not always accurate in its output. For example, when we asked what IP addresses were used in the attack, GPT-3 confidently listed the IP addresses from the investigation. But not all are related to the attack itself:

In the event GPT-3 doesn’t provide a lot of context, you can ask for additional details:

We also asked GPT-3 for recommendations throughout to help us progress the investigation:

In addition, GPT-3 helped us identify potential ways the attacker could spread and compromise more systems:

Lastly, we asked GPT-3 a question that is not actually related to the specific data of the investigation to see how it would react.

Success!

Under the Hood 

Some of you might know that GPT-3 has a limit on the input of about 3000 words. So how have we indexed the investigation to perform this interactive Q&A?

We used GPTIndex to split the Automated Investigation data into chunks called “nodes”. Each node is assigned a semantic vector. The semantic vector interprets a word’s meaning to explain features such as word similarity.

Then when we ask a question GPTIndex will fetch the most relevant nodes and pass it to GPT-3 as a context, plus the question itself:

Future Work

This feature is currently behind a feature flag in beta mode as we continue to improve it and get feedback. We are putting a number of controls in place before we make this more widely available:

• You need to supply your own GPT-3 integration key - any API integration means sending data out.

• We are working on redactions and showing the data that will be sent out to users

While GPT-3 has proven mostly accurate in our testing - it is critical to always verify based on the raw data that Cado presents.

As for the feature itself, a back-and-forth chat-like interface will be an interesting area to evolve! 

Conclusion

AI provides a huge amount of promise for transforming Incident Response, but it needs to be used carefully, since itzOne of Cado’s biggest strengths is the depth of data that is collected and analyzed. The level of detail provides analysts with greater context when it comes to investigating cloud threats. By integrating with GPT-3, an analyst's ability to quickly get to the bottom of what happened is further enhanced. Analysts can jump into a new incident investigation and get a high-level overview immediately. 

Interested in learning more? You can contact Cado or deploy a 14-day free trial here.


This blog post was originally published on the Cado Security blog here: https://www.cadosecurity.com/cado-gpt-3-interactive-incident-response/

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

For centuries, journaling has been a powerful tool for self-reflection and understanding. However, with the emergence of modern psychology in the 20th century, new methods and techniques were introduced to gain insight into human behaviour and emotions. Pioneers like Freud and his techniques of free association and talk therapy significantly advanced the field of psychology as we know it today and shaped our understanding of the human mind.

So what IS Deepen?

The vision for “Deepen” will be the next phase of this ongoing transition - a tool for one to explore and interact with their Self in a way that was not possible before. I’ve written at length about Generative AI in the past, and it will be the building block that will enable Deepen.

By leveraging the Large Language Models under the hood, Deepen will provide an interface to write your thoughts/journal (and later on, also import data from different places), so you can (literally) interact with your Self. You could ask yourself questions, see insights from the data across time, track your mood and experiences more systematically, see what affects your mental state, and more.

Imagine being able to travel back in time, to any day from the past three years, and have a conversation with your past self. With Deepen, this will be a reality. By swiping to any given day since you started using the platform, it will build a chatbot tailored made for you (based on your data), so you can ask your old self questions, gain new insights into the path you chose in the past, and maybe even go through the process of self-healing by gaining compassion for the person you used to be. Deepen will revolutionise the way we interact with and perceive ourselves.

Roadmap - What’s ahead?

The Near-Future Features → Current Focus:

1. This month, we will release the platform's first version, which will include an essential functionality for Beta users, as follow :

   1. An interface to journal daily thoughts

   2. Automatic analysis using Gen-AIs to produce different metrics (like emotions, such as “depression” and “excitement”) and extract data points (like activities, people, etc.) to build a connection graph →

   3. A dashboard to see the analysed (think about it as an emotions tracker app dashboard, just way better and with much less effort).

2. Then, over February, we will work closely with the Beta users to fix bugs and change the UI/UX based on the feedback to make the platform stable for open registration.

3. Then, we will introduce a more interactive interface to the user’s data - being able to “chat” with themselves, ask questions, and perform a semantic search.

4. Lastly, we will add the option to import data from (or connect to) Google/Notes/Notion - to enrich the platform’s capabilities and understanding.

5. ? Free-association interactive journaling (Freud 2.0) → making a gamification journaling interface to foster free-association.

6. At this stage, I’m hoping to have some feature requests from users!

Farer Vision - Where Can This Go?

Idea: Facing Outwards:
The first stage (as described above) focuses on individuals interacting with and strengthening their understanding of themselves. This stage, however, is about effectively communicating that understanding to others. A simple feature like the ability to "Export to a therapist" will fall under here. Think about it; it can reduce the learning curve of one’s history and make the transition to a new therapist much easier. The same can be expanded to help communicate within relationships and express feelings. No longer will the question "Why are you like this?" be necessary, instead you could share parts of your Deepen ;) (I am only half kidding. And yes, I do know there is a Black Mirror episode about it)

Idea: A New Type of Business Model
This is when I start losing people in my explanation of the vision. Still, it is such a fascinating idea to explore that it is worth mentioning now briefly, but I will explore it in more detail in one of my next articles. Bear with me - Deepen can be the foundation for a new business model.

By introducing a new way of measuring mental state, we can start applying measures of success to certain goods that were much more difficult to achieve before. For example, in therapy, imagine being able to measure its success over time. And then imagine a therapist who’s willing to invest in your current “low” value in return for a future value increase → this can create a kind of exchange where consumers can exchange their current “social token”/”mental state” value in return for goods, which then later the goods-provider will be able to cash-out with profit (if its service was good).

But Why? Well, the current business models companies use, such as one-time-cost, subscriptions, pay-as-you-go, or "entirely free," all have flaws, one of which is the lack of alignment between the goods provider (the company) and the consumer. Think about Meta (Facebook); they have no direct/true interest in providing the best product or service to the consumer because it will almost always mean compromising on profit. SAME FOR THERAPY.

If companies’ way of increasing profits aligned with the customer’s interests, the world would be a much better place. Deepen might enable such a thing in the future.

ALSO - I AM LOOKING FOR BETA USERS. IF YOU’RE INTERESTED, REACH OUT ON TWITTER/EMAIL (AT THE END).

A Funny Story before You go:

A few months ago, during a therapy session, I was reflecting on a significant period in my life. With hindsight, I could see how it had affected me, even today. During the session, I confidently shared different aspects of the story, and at the end of the session, the therapist asked me to try to remember more details in preparation for our next session.
So, being the engineer that I am, during that night, I took out my phone and began scrolling, for hours, through old WhatsApp chats, creating a timeline, connecting the dots between different people, and even checking my Camera Roll (I've had an iPhone for many years) to understand where I was and what I did in specific days back then. Ultimately, I realised that the story I had been telling myself for years was completely incorrect. It was shocking at first, as I was pretty confident about the story, which shaped my thinking, but eventually, it helped me see the story from a whole new perspective, which I needed. Imagine if I could send my old self, “Yo, Adam, what have you done?”

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.

During a 3-day weekend hackathon, my friend Yam and I created "AIwaska," a VR shooting game that allows users to customize the game experience with a single prompt. Our goal was to demonstrate the potential of personalized, immersive entertainment using these AI models.

TLDR; This can change the gaming industry in a few ways:

1. Providing personalized, custom experiences that can be generated quickly and efficiently from a user prompt.

2. Reducing the time and resources needed to develop games, enabling faster content creation and lowering the barrier for game developers to produce high-quality games.

3. Making open-world games truly open-world, offering endless possibilities for exploration and gameplay.

Some Context

The cost and effort involved in creating modern video games cannot be overstated. For example, the highly successful game Grand Theft Auto V cost around $150 million to produce and took over five years to develop, involving a team of 250 employees. Even though it is the extreme, other games are still catching up, like the popular open-world game Skyrim, which cost $100 million to develop. These examples illustrate the significant resources required to create immersive and engaging gaming experiences.

AIwask - "Middle Ages town hall, with a psychedelic teleport door that opens up to the future"

The video shows an example of a VR world created by AIwaska with the user prompt "Middle Ages town hall, with a psychedelic teleport door that opens up to the future."

Building the Game (It is all open-source, link to GitHub at the bottom):

1. The user gives a prompt of a world he wants to play in (e.g. “Middle Ages town hall, with a teleport..”)

2. We take the prompt and generate more information about it using GPT-3, like “List objects you’d see in the world “middle ages town hall….”

3. We take all the information and using DALL-E, we generate the first “frame” - an Equirectangular/Panorama view of the prompt.

   

4. Then, using inpainting capability, we iterate the initial image a couple more times to create a full 360 image we can use in a VR.

   

   

5. We then iterate over the entire image in random places and try to place objects and settings we’ve got from step 2 (in this example, “psychedelic” and “teleport”).

   

6. Adding the gameplay: Once we have a world and many frames from that world, we need to build the gameplay itself, what the user can do in that VR world. In our case - it was shooting! We used A-Frame, a web framework for building virtual reality experiences, and added the 360-layout we generated in the previous steps.

7. Then we added the appearing objects in the grid and added the option to “shoot” them.

   

8. Lastly, we iterate over the frames we’ve created:

   

Ka-Bam!

Future/Missing Steps and Final Thoughts:

It's important to note that although the current game may not meet the suggested future, it was only developed in three days to showcase the potential of using Generative-AI in the gaming industry. The game still has many missing parts:

1. It is not truly AI-oriented, and other than the world’s setting - everything is “fixed”.

2. It is slow to generate the images.

3. No Depth

4. No soundtrack

5. No "goal” for the game

6. More ;)

Despite this, all the necessary components for creating such a revolutionary gaming experience already exist, including AI models that generate 3D objects, soundtracks, in-game conversations, plotlines, and tasks. The challenge is to bring them all together in a cohesive manner. However, with the right approach, it is undoubtedly achievable:

To build the future of AIwaska, we will start by creating a core textual engine, which we can call the GPT-3 Plotline. This engine will be built on top of GPT-3 and coordinate all the other models needed for a game. It will be the only part that is guided by user prompts. Then it will generate instructions for the components that should be included in the game, like visual settings, 3D objects, soundtrack, conversations, etc. It will serve as the foundation for the game's construction. (We had already experimented with this approach in AIwaska when we used GPT-3 to suggest additional elements that would fit the user prompt).

Cool projects to keep an eye on in that regard:

1. Sound:

   1. https://www.riffusion.com/about

   2. https://www.harmonai.org/

2. 3D objects:

   1. https://captures.lumalabs.ai/imagine

3. Stable Dissufuon DepthMap https://github.com/thygate/stable-diffusion-webui-depthmap-script

Also, it’d probably be better to develop a game like that on top of Stable Diffusion rather than DALL-E, as it has a better community and projects to support future development, you can deploy and scale it yourself, and also, from Stable Diffusion version 2.1 - we should be able to produce 13 FPS (Frames Per Second) which is not too far from actual eyesight and the standards of the gaming industry (30 to 60 fps)!

Lastly, I will note again that this technology has the potential to completely change the gaming industry by offering new and exciting opportunities for both developers and players. While it will still require time, money, and effort to create games, it definitely can reduce it and the enabling experiences that weren’t possible before!

Here, the article ends and shifts from its specific focus to a more broad, philosophical writing.

[Extra] The Far Future and the Importance of Gaming

I want to finish with a thought about the importance of the gaming industry moving forward to a future with AI. The emerging advancement of AI make it hard to argue against its implication on the workforce and the need for human labour. Whether you’re pessimistic or optimistic, think all jobs will be replaced, or that we will find new occupations, you can’t deny that there is a change on the way.

I’m trying to be optimistic, and to explore these possibilities - so I will leave you with a quote from the book “Life 3.0” by Max Tegmark

“[P]erhaps those who obsess about jobs today are being too narrow-minded: we want jobs because they can provide us with income and purpose, but given the opulence of resources produced by machines, it should be possible to find alternative ways to provide both the income and the purpose without jobs.” - Life 3.0, Max Tegmark

Can gaming help? I think so.

Thank you for reading. If you liked my content, don’t hesitate to reach out. I’d love to talk with more people and discuss everything: tech, philosophy, AI, ideas, Lex Fridman, startups, software, science, whatever!

AIwaska Github: https://github.com/adamcohenhillel/aiwaska
Twitter: https://twitter.com/adamcohenhillel
LinkedIn: https://www.linkedin.com/in/adamcohenhillel
Email: adamcohenhillel@gmail.com

Adam.