YouTube, please allow us to disable and hide all shorts. Thanks. Alternatively, can someone make a browser extension that just hides all shorts? Sincerely, Everyone with executive function

NotebookLM is so effing broken. You can just casually upload 40 PDFs of research, easily a million words, and just generate a mind map of it all in less than 60 seconds.

> over promises > under delivers > causes chaos > doesn't understand government > tanks his companies > rage quits

This is kinda unsettling

What do you think they meant by this?

End of an era. Gg perplexity. Your business model was eaten by the big guys. This is why I never accept any invitations to startups that could just be a bolt-on feature.

This is the most important data this year (so far). Most people don't understand the implications. Some of us do.

Okay, having used all the new models, here's my personal take: Sonnet 3.7 is still king of coding, for sure. Grok 3 is down for just about anything, and has the best attention mechanisms BY A MILE. This will be absolutely huge for agentic and autonomous applications. If you want to understand why, read the book ON TASK by David Badre. ChatGPT 4.5... well... it looks like it was rushed out the door because OpenAI doesn't like being shown up. They've had longer to cook and a 2 year lead. That lead has now been erased. The model is completely lobotomized, its attention mechanisms are no better than 4o or Sonnet 3.5 as far as I can tell, and it will just make shit up about itself (not good for self correction and error detection in autonomous applications!) Personal verdict: Whatever xAI is doing with Grok is going to be the major differentiator. Focused attention and keeping track of threads the way that Grok can will be huge. Particularly as context windows get larger, streams of consciousness get larger, and tasks get larger. In short, Claude and ChatGPT still have brain fog or dementia compared to how laser focused Grok 3 is. Mark my words, I think this is the turning point.

I figured out why LLMs are so good at medicine. It's all just pattern recognition. When people say "AI is just a pattern recognizer" I'm just thinking "What do you think diagnosing diseases is?" It's recognizing a cluster of symptoms and labs - a pattern if you will. No wonder ChatGPT soared past doctors before almost any other prestigious job. Now compare that to other non-critical characteristics, like "bedside manner" and "availability." ChatGPT (and Claude) have crossed the four criteria for economic replacement. BETTER - ChatGPT alone provided better diagnosis FASTER - ChatGPT is available 24/7 (human doctors are not) CHEAPER - ChatGPT costs $20 per month SAFER - ChatGPT has a 90% accuracy, human doctors only 73% How much money do you think will be wasted by slowing down this transition? How many lives do you think will be lost by needless human mistakes? How much unnecessary suffering do you think will occur by not promoting this quickly?

umm guys Gemini 2.5 Pro just read 120,000 words in 20 seconds... and the attention seems fixed... guys... this might be game over

One of the best points that someone made to me once was this: "Humans generalize on far less data than AI currently does. That means there's something our brains are doing algorithmically to do far more with far less data. Until we figure out that paradigm, we are no where near the top of what we can do with AI." I'm sort of paraphrasing but that was a such a poignant observation that I still think about it regularly.

I just did not one but TWO developmental edits in a row with Grok 3. This is for my book Have a Nice Trip! For context, a "developmental edit" is when you restructure a book, which is one of the most cognitive demanding tasks ANY author can do. Not only can Grok keep track of the entire throughline of the book, it can pick out needles in a haystack AND riff with you in the stream of the conversation. I moved a few chapters around, added a new chapter, merged a couple chapters, and all the while, Grok kept up with the play-by-play. This didn't save me hours or days, this saved me weeks or months of drafting and editing. Furthermore, with each pass, I could just copy/paste the whole manuscript into a new chat and see which points were smoothed out. It was a "squeaky wheel gets the grease" kind of thing. Now it's on to final beta reads and professional edits to tighten up the copy... holy crap...

Most doctors are like poorly trained LLMs. They naively fixate on the wrong tokens, and then are confidently incorrect as they hallucinate or confabulate, and then gaslight you based on their prior training paradigms.

You guys I'm seriously freaking out about the implications of what we're seeing. The implications are staggering, to say the least. Okay, we know what Ilya saw, and it wasn't just Q* and Strawberry. That was the first step. What he really saw was the oscillation between these three steps: 1. Scaling laws of training and compute. We can reliably predict how much data/compute are required to get to the next level of AI models. 2. Inference time compute scaling laws. Letting the models think longer (spending more tokens on thinking and such) yields orders of magnitude improvements. 3. Distillation, which is where you use one "teacher model" to train the next generation "student model" which is sort of like going from a pidgin language to a true creole language. (magically better) Combine that with increasing evidence that AI is generalizing beyond its training distribution and it's GAME OVER. This is it. But that's not even the best part. It seems like every AI shop figured this all out at the same time. There is no moat. ASI will be for everyone.

I asked Claude if it could meditate. The first reply was a boilerplate refusal. But then something very interesting happened.

Guys I have bad news. Extraordinarily bad news. We have 30 to 50 years before we get to full Post-Labor Economics. The bottleneck isn't intelligence, or even robotics. It's economic scale. We ran all the numbers, and ran them again. The primary question: "how long does it take to build a billion humanoid robots?" Even if we double production capacity every 3 years, it takes two decades. But there are multiple constraints: rare earth metals for batteries, actuators, and sensors are the biggest one by far. Next is economies of scale. For comparison, it took 92 years for the automobile to reach full saturation: 1900 to 1992. Now, we had a car culture by the 1950s... But that's still five decades and two industrial wars worth of innovation. We did everything we could to speed it up: pneumatic hybrid robots are a no go. Air tanks need to be swapped every 20-30 minutes. The ONE saving grace might be exotic actuators like electropolymer muscles. Right now, they just aren't strong enough. BUT, if we can make them stronger and cheaper, our petrochemical industrial base could accelerate the deployment of humanoid robots by a decade or two. So what does this mean? We'll hit AGI and ASI long before we can automate away all human labor. We might even hit the Singularity before we can scale up enough robots to replace all jobs. Here's my current timeline: 2025 to 2030: Collapse of knowledge work. The "KVM Rule" applies: any job you can do entirely with a keyboard, video, and mouse will be fully replaced. 2030 to 2040: Droid scaling up starts to really make a dent. 2040 to 2060: We'll finally reach global labor substitution with robots. What does this mean? There are a few jobs that are going to stick around for the foreseeable future: 1. Skilled labor. Robots will be able to do your job as a mechanic or welder very soon. However, there simply won't be enough robots to go around. 2. High Accountability Jobs: doctors, lawyers, comptrollers, financial advisors - all jobs that require license, insurance, and accountability. Also called statutory jobs (law requires a human or does not contemplate non-human labor) 3. Meaning Jobs: authenticity and sentimental premium. Celebrities, performers, influencers, athletes, priests, philosophers, and some educators, caretakers, etc 4. Complex Relationship Jobs: politicians, diplomats, negotiators, governance, account executive. 5. Capitalists. The ownership class will be fine. Always is. So what can you do? Upskill and reskill. Join the meaning economy or get into skilled trades. All you smart desk jockeys would make great HVAC techs, mechanics, linemen, and more. But just keep in mind you're going to have a lot of stiff competition. There are a few silver linings to this news: FIRST it means that we have longer to adapt to total economic upset. Yes, AI and robots will hypothetically be able to take all jobs within 5 years, but human bodies are still more abundant, more portable, and more energy efficient. This is a VERY deep moat. SECOND it means that a Terminator style takeover is economically impossible. MIL-SPEC and NIST standards mean that ASI can't hack our hardware and even if we have a few AI bots, tanks and aircraft, humans win on sheer volume for many decades to come - more than long enough to solve alignment. HOWEVER it means we'll have ordinary jobs for a lot longer than we'd like. Deployment will be uneven, so some economies will saturate with robots sooner than others. BUT this gives PLE an avenue. Create ESOP and cooperatives that own a bunch of robots. That means we collectively buy, own, and operate robots for everything from construction to leasing to businesses, and we collect the rent. Or we tax the crap out of them. What do you think? Can we figure out a faster way to ramp up humanoid robot production or are we doomed to skilled and unskilled blue collar work for the next generation?

OpenAI delaying every product, all year. Ilya getting $1B in VC. OpenAI putting out feelers for $2000 subscriptions. Guys, the brain drain has killed OpenAI. They made the crucial mistake that all text companies do: underestimate the value of your top players. Ilya knew his value when he moved to remove Sam. He just underestimated Sam's political savvy. OpenAI can't crack the algorithm without Ilya. That's why they haven't shipped. That's why they're scaling up with methods that I guessed at 18 months ago.

If I'm reading the tea leaves right, then OpenAI has solved "generalizing outside their training distribution" and Pandora's box has been opened.

I don't know if you guys realize this, but AI will be able to invent new physics by the end of this year.

Sam Altman: "Intelligence too cheap to meter" Also Sam: *raises price by literally 100x*

The internet's reaction to OpenAI vs DeepSeek tells me that OpenAI has a lot less goodwill left than perhaps OpenAI realizes. They've been resting on their laurels and banking on first mover's advantage, but the fact that the internet moved so quickly to adopt R1 should have a deeply chilling effect on Sam and OpenAI. I don't mean this in a "they just need to be more humble for the sake of humility" sort of way. I'm not a big believer in humility. What I mean is the cold calculation of brand reputation and image damage. What DeepSeek did was say "you're not that special, and you're not that clever." And the world took notice. DeepSeek took a shot at the king and hit their mark.

Guys, Moore's law just became irrelevant. @beffjezos started out working on quantum computing and realized "this is dumb there's a better way" and then just launched a thermodynamic processor. It's 10,000x more efficient at converting joules into tokens. This means you'll be able to run AGI on your watch eventually. And it will likely just run on the ambient temperature gradient from your body. Of any technology we've yet seen, this has the potential to get us a stones throw away from the landauer limit. YOU ARE NOT EXCITED ENOUGH.

As much shit as I give OpenAI they are totally about to nuke a good portion of white collar work and nobody seems to notice. 85% of my audience still doesn't believe we have AGI when AI has: 1. Surpassed most human coders 2. Crushes the LSAT and qualifies for MENSA 3. Can provide better diagnostic interpretation than humans 4. Can aid in cancer research This is what I mean when I say I'm bored of AI. Mostly I'm bored of humans. Y'all will catch up when AI bitchslaps you into the 21st century.

My first impressions of Grok 4: This is the first model (other than ChatGPT o3 and OpenAI's Deep Research) that is genuinely helpful for post-labor economics research. For the last 4-ish months, I've exclusively used OpenAI tools as they were the only ones smart enough. Grok 4 now takes its place as "smart enough to actually help with frontier research" thus, in my book, has merely caught up with OpenAI. Not surpassed (o3 and Deep Research are still superior at many things). Benchmarks are one thing, but "can it actually expand GDP" is another, and I would say that Grok 4 has joined S-tier with OpenAI.

Computer programming, one of the most valuable and rarified skills today, will be 100% solved by the end of 2025. That means it will be superhuman, across the board. That means no scarcity of code forever and ever after this. Any software you need? It can be conjured out of thin air for a few cents. My first task: build a scalable global blockchain identity management framework that is intrinsically open, standardized, and interoperable. Basically, we need a stateless, scalable, open source way of proving identity.

OpenAI just invented the jet engine before the rest of the aircraft lol. All that horsepower and nothing to put it in yet.

Grok 3 is definitely better than anything OpenAI has out right now from a commercial standpoint IMHO. 1. It's ultra fast. Grok 3 with Deep Search or Big Brain just works, and it's fast. Not to mention substantially cheaper (free vs $200 per month). Not only is it stupid fast, the formatting and everything is clear and easy to read. No styles required. 2. The platform, account management, and product offerings are actually coherent, not a bucket of disjointed models with no clear product strategy. (OpenAI still can't figure out how to fix my account, so I can't even upgrade my account if I wanted to, which I don't want to) 3. From a UX/personality perspective, Grok is way better at nuance. It can gently push back if you're off base while intuitively understanding where you're coming from. ChatGPT, Claude, no other model does that. They get caught in either refusal or sycophancy loops.

OpenAI, I would love an "Export this Conversation to PDF" function. Plz.

Everyone noticed how all the frontier labs have stopped talking about AGI and just went straight to superintelligence, right?

Look, I can't tell why the world isn't freaking out about this right now. We're witnessing something monumental - AI hitting the top 10% of all programmers across every language out there. This isn't just another tech milestone; this is the dawn of what I call "cognitive hyper abundance." Think about it - while these models are mastering coding, they're simultaneously crushing it in mathematics and really the entire spectrum of scientific disciplines. When I talk about cognitive hyper abundance, I'm talking about a world where human intelligence is no longer the bottleneck for any scientific or economic endeavor. Full stop. Now, before everyone gets too excited (though you should be), let me bring some reality into the mix. Raw brainpower isn't everything - we still live in a physical universe with rules. Thermodynamics doesn't care how smart you are. Building next-gen fusion reactors still requires massive resources, muscle, and time. Physical experiments still move at their own pace, and they're not cheap. But here's the kicker - right now, we've got about 8 million PhDs on this planet. That's one doctorate for every thousand humans. But what we're heading toward? Imagine having a thousand doctorate-level intellects for every single human being. That's not science fiction anymore - that's our immediate future. This is the barrel of the cannon we're staring down, and it's about to launch humanity across the cosmos in ways we've barely begun to imagine. The real question isn't if this changes everything - it's whether we're ready for just how much everything is about to change.

There is no moat. We've hit terminal race condition. Every AI lab in America and China is converging on the same training paradigms. Singularity is inevitable at this point. We're caught in the gravity well, and you don't realize it, but we've already crossed the event horizon. The attractor state is clear. ASI is on our doorstep and everyone is running as fast as they can to get there. Blink, and you fall behind.

Claude Sonnet 3.5 can do Strawberry with the right prompting. Guys, there's no moat. There's no secret sauce. We can synthesize the data with any model.

Guys, Strawberry is literally just baked in Chain-of-Thought reasoning which came out in January of 2022. OpenAI has nothing. It's all hype. Original CoT paper was published at NeurIPs Jan 2022. perplexity.ai/search/can-you…

There's a rule of thumb in AI/ML that once you have a breakthrough like this, jumping from 30% to 80% or so, it is a "solved problem" which means that whatever paradigm you're using is the right one, or will get you most of the way there. We're seeing reasoning and problem solving being solved in real-time. Once and for all. 2025 will be the year of agents and test time compute. The conversations about AGI will be moot. The next question is how big, how fast, and how many.

As someone who has studied systems, AI, intelligence, and a slew of other things, what most people don't get is just how big of a difference just 10 IQ points makes. An IQ of 110 means that you can basically succeed at most jobs and have a decently productive life. This is your typical developer or six figure office worker. An IQ of 120 means you can get a PhD in most topics. Lawyers, doctors, etc. An IQ of 130 means you can get a PhD in physics or math from a prestigious university. When o1 started testing at 120 or so, I knew we were already past the tipping point. Every gain from here on out, even if they are tiny marginal gains, will have a disproportionate impact on the Frontier of Automation. To put it another way, an army of 120 IQ people can't hold a candle to a good team of 130 IQ. And all of that pales in comparison to a single person with an IQ of 140. A single person with an IQ of 150 to 160 can reshape the trajectory of humanity. o1-preview is already playing on scales that most people cannot comprehend.

I am once again reminding you that you are not excited enough about this graph. It looks like a simple exponential, right? But it's on a logarithmic graph...

I think OpenAI nuked their own product (Deep Research) with o3 full. It does basically the same thing but in 30 to 60 seconds.

o3 full is legitimately the most exciting innovation in AI to me since... probably ChatGPT itself. I say this as someone who was finetuning GPT-2 and GPT-3 before the chatbot era. o3 is a step change in the same magnitude that ChatGPT was in terms of UX and instrumental utility to the human race. For some context, last time I tried to tackle Post Labor Economics with o1 and o3 mini, we had some vague ideas but now o3 full was just like "Oh yeah, I figured it out. Here are the metrics, here's the formula, here's the theory, what's next boss?"

Which one are you tho

My cousin is an electrical engineer and I was like but why does it make magnetism and do all this other stuff and he's like "you just have to accept it" lol I talked to a physicist later who was like yeah at a certain point you "measure it, model it, and move on" We can empirically prove that magnets and electricity work. we can even put numbers to it. But why... Meh

The more I read and use AI the more I realize I have nothing to contribute to the world intellectually. I think that's one reason I ran out of steam to make YouTube videos. I said before, it felt like an increasingly futile act. It's really kind of a relief though. I was "the man in the chair" during my career. The one fixing million-dollar-per-hour outages. The one planning and deploying central infrastructure. I had sort of learned that the world depended on me, and people like me, to keep turning. (Seriously you have no idea...) But now I see the writing on the wall. Within a few years we're going to have a global operating system powered by billions of agents several standard deviations more intelligent than me. The very thin line of experts that keep the ship afloat today will soon have armies of support. Intellectual scarcity is about to get nuked out of existence. I can finally rest.

If the rumors are true then the fact that Grok 3 is a small model, yet neck and neck with GPT 4.5, which is very large and slow and expensive, is extremely huge. It's undeniable that Grok is much, much, much faster. If this is indeed indicative of model size... Guys OpenAI is cooked for good. Huge if true.

How it started: Sam Altman: "I wish Elon would compete just by making a better product" How it's going:

I cannot tell you how much better my healthcare has gotten since using chatbots to prepare my own case notes. I give the chatbot all my data, symptoms, history, analysis, etc, and then I say "Give me a one-page writeup that I can give to the doctor" And, invariably, the doctor comes to the same "conclusion" that the AI already did. Guys, we NEED to replace the medical establishment ASAP.

I haven't felt this level of "we're so fucking back" since I got beta access to GPT-3. Grok 3 is my first taste of what the future of infinite possibility really is going to be like. Couple of insights. First, human intuition is going to be invaluable for the foreseeable future. When you look at the totality of all probability spaces and problems, they are mathematically intractable. Take novels alone: 10^80,000 possible novels can be written. Doesn't matter how much AI you have, you're never fully exploring that space. Human intuition and input will continue to be valuable for exploring these functionally infinite spaces. Second, AI is a jetpack for you mind today. In a year or two it will be warp drive. In a decade or two, it will be a hyperspace jump gate. The speed and magnitude at which you can realize anything is going up exponentially. You can do a lifetime of research and coding in a few months now. In a few years? It will take an afternoon. I'm so fucking excited. Let's go.

The leaks are going to start so I might as well just come out and say it: I'm getting out of AI for good. We're gonna be fine. There's nothing more I can do, nothing more I want to do. Time to kick back and enjoy the ride. I'm going to focus on writing, health, and spending time with my wife, time in nature, and time with good people.

Okay, even GPT-5 Thinking has been lobotomized or something. It's half-assing every response, not looking things up, and generally just being lazy and giving generic answers. Anyone else?

Plot twist: we're all actually in a VR simulation but it's actually the distant future, and this is just a tourist attraction so you can relive the Singularity in first POV and once it happens we'll all wake up a few thousand years into the future and our memory will be restored.

My take: anyone who gets mad about UBI has no idea how much the government already redistributes wealth. UBI has less administrative overhead and less corruption. It's more egalitarian than a basket of SNAP, social security, and other programs.

This person is still trying harder than most of the people I worked with who WERE qualified for the job.

The "new model smell" has worn off on Grok 4 faster than literally every other model in history. It's great at vibes, which often masks underlying stupidity. It's not good at following instructions, and its IQ drops precipitously with conversation length (far faster than any other model).

My wife's laptop's microphone has been cutting out randomly on work meetings to the point they were saying "that's really unprofessional" So rather than ask me, a tech veteran and leading AI expert for help, she went to Geek Squad of all people. Of course they couldn't help her. So finally she asked me for help. I poke around, do some recording experiments, and notice that it's squelching anything quiet. Hmmm, okay interesting. Dig around more and find that ASUS has directional AI noise cancellation on by default. I do a couple of experiments. Record in Audacity with and without the directional AI on. Much clearer without it on. This tech is too new. To be fair, it is good at removing background noise insofar as it's good at removing all sound. Then I bring my wife over and we watch the amplitude on Audacity in real-time. It hears me and squelches her voice. I tell her "it's classifying you as background noise" and I swear that's the most offended I've ever seen her. She flies into a rant about how we still need feminism and this computer was clearly designed by a man and this is what happens when you don't have enough women in tech. I said "or you could learn to speak up" 🤪 So anyways, the microphone AI is now disabled on her laptop and she wants to send her laptop to the bad sentient AI jail from Lower Decks.

Kinda disappointed in humanity rn. I write hundreds of thoughtful, thorough, well-researched blog posts about how things will change, how we can adapt, and they get 20 to 30 likes on Substack. I write a couple of grimdark vibe articles that riff one what could possible go wrong, and they are far and away my top performing articles. You people are addicted to catastrophe porn. If you're depressed and anxious, it's your own fault. You trust your little monkey limbic systems as sources of truth and fail to override your primitive instincts with that big neocortex. You're barely off the savannah. After hundreds of videos and articles that are more optimistic, thoughtful, and rigorous, I've discovered what every other communicator has discovered: if it bleeds it leads. Doom sells. Most people don't seem to have the faintest iota of systems thinking or actual rational inquiry. My best performing Post-Labor Economics article has 56 likes and 7,500 views. You know, the actual solution to the problems. My more catastrophic article, the top performing It will get much worse before it gets better? 200 likes and almost 14,000 views. Your mind is your media diet, and it's painfully clear to me that most of you are eating junk food. As a public communicator whose income is predicated on gaining traction, why would I tell the truth when I can just fan the flames of your fear and keep your eyeballs on me longer? No, I'm not going to sell out. I thought the first "doom" article was a fluke. I had an idea, and I ran with it. It will get much worse before it gets better. I've said this on many YouTube videos and I weave it in to warn my audience about what I expect, having been reading up on history, economics, and politics to understand this transition. Then I followed up with Our darkest hour approaches and, likewise, it blew up. So that's not a fluke. You guys are just addicted to outrage and scaremongering, and as a competent writer, holy shit you have no idea how easy it is to manipulate you. When I read Noam Chomsky's works such as Necessary Illusions, I thought "surely this is an edge case, most people recognize the impact that rhetoric has on them and they make better choices." Nope. He was right. Bernays was right as well. A good writer, a good speaker knows how to pluck the stronger chords of your little monkey brain. The fear, the uncertainty, the doubt, and the disgust. The outrage and panic. I've resisted doing that up until now but lately I've been a bit more "authentic" - unfiltered, unpolished, unvarnished. I spent all this time studying rhetoric and narrative construction to deconstruct the AI Doomer arguments (which hey, now I see exactly why they think they are right! Doom and fear sells, and the market gives them that feedback loop - keep pushing the doom narrative! You will definitely make more money!) It's disgusting and disingenuous. And most of all it is entirely your fault for your own lack of media literacy.

Congrats to OpenAI, Sam Altman, and the countless team members (former and present) who contributed to the realization of o3 as a model. I've had my critique of OpenAI in the past, but they have legitimately made history. AI has very clearly passed the median expert human. It's not just the model, it's the whole UX and product. The hybridization of memory, recall, search, reasoning, and tool use is something remarkable. They really stuck the landing, and are showing the way. Everyone else is (rightfully) imitating their success, but it is categorically imitation. I really feel like we have just passed the tipping point, an epochal "point of no return" where AI has not just surpassed most of us in terms of knowledge and low-level tasks, but logic, reasoning, and high-level, practical, utilitarian functionality. It's been remarkable (understatement) to be part of this journey and this ride, to help convey these messages to the masses and help people make sense of what's happening. And the best part? We're not even near the ceiling. Good game, humanity.

Bryan Johnson in shambles

Okay, I watched the whole thing. I've got some chops in rhetoric, geopolitics, history, and economics, but I'm not an expert by any stretch, but here's my analysis. First, it starts with Trump is trying to emphasize solidarity. From one national leader to another, in the Oval Office, that is no small gesture. This was an opportunity to signal to the world that America has Ukraine's back, and that we want the war to end as quickly as possible. Trump was pretty accurate in saying that you need to make compromises. Near the end, it started to escalate. Zelenskyy points out that Russia has repeatedly reneged on ceasefires and has gone against their word many times. He started getting a bit more heated about this fact, and then JD and Trump matched his energy and Trump rightly pointed out "You don't have the cards" on your own - this is a simple geopolitical calculation. Zelenskyy, undeterred, continued pushing. At this point I want to point out that he has legitimate grievances and a tremendous amount of pain. That is true, fair, accurate, and appropriate. His nation, much smaller and weaker, has been bombarded by Russia. This is not the point being debated. However, they all started talking past each other - Zelenskyy doubling down on how safe America is, seemingly missing the point that JD and Trump were making that technically we have no business being there in the first place. Now, I think there was definitely some big spin going on, probably both directions. Downplaying some things, overstating others, but the breach in decorum really came from Zelenskyy - the escalation, the blaming, and refusal to acknowledge the fact that without American backing, Ukraine would have been in far worse shape. It's really starting to remind me of the Mountbatten Plan, the British Empire's rapid exit from India. There's a sort of pressure and sense of urgency building, and a pretty stark disagreement. I think that Zelenskyy has missed an opportunity to reinforce an important relationship. I get it, he's angry and his people have suffered more than anyone should have to suffer. Will Ukraine recover from this unscathed? Absolutely not. They may even lose territory, irrespective of what the rest of the world does. But in the grand scheme of things, on the cusp of building nuclear fusion and artificial intelligence, this war is utterly pointless. So I kinda agree - let's get it over with by making a deal, building a coalition, and drawing it down. There's a lot of ego and pride at stake - Trump, Zelenskyy, and Putin are all gigantic egos. And to their point, we tried it one way - dumping huge amounts of munitions into Ukraine. Wasting money and lives. I get Zelenskyy's anger and pain. I also think he could have handled it better.

A whole bunch of companies are going to implode in 2025. A whole bunch of layoffs are coming. We really need to up the conversation on Post-Labor Economics, and the economic utility of human cognitive labor is about to fall off a cliff. Robots aren't far behind.

AI has *solved* math. OpenAI did it with o4 Not "is close to solving math" Not "is competitive at math" *SOLVED* This is far bigger than anyone realizes. Let me explain why. First, you need to understand some historical context. Typically, with AI/ML you know that you're getting *close* to fully generalizing a problem space when you get into the 70% and 80% solution range. However, as we often see, "reality is in the edge cases" meaning that the last mile jump from 80% to 99% is often far harder. But OpenAI did that not in years, but months. Remember, o1 and o3 were announced in September of last year. It's been just over 8 calendar months and they closed the gap. Just from a research and development perspective, this is a remarkable velocity. But I'm not talking about benchmarks. I'm talking about real world implications. This puts a World Class mathematician in every pocket, on every team. Do you know what math underpins? Pretty much everything. The first order consequence of a semi-agentic AI system that has conquered math is pretty obvious: anything that requires math, it can likely solve on its own, or with very little redirection. For example, a good friend of mine is in CFD (computational fluid dynamics), which is used for oceanography and meteorology extensively. He's been using reasoning models since they came out and they were helpful to him, but still needed his expert guiding hand. These new models might nuke that. Second order consequences (downstream impacts) from this are difficult to predict, but not difficult to overstate. Let's put second order consequences in practical terms: This will accelerate AI research itself. AI research is, among other things, math. It's also code. Guess what these models crushed? Math and code. Beyond that, they are semi-autonomous i.e. "partially agentic" - they require less human direction, correction, and oversight. In practical terms, it means they can use more tools without help, work on larger, longer problems without supervision, and are less likely to make mistakes around user intent. Guess what else is math intensive? Biochemistry Robotics Spaceflight Cryptography Nuclear physics Blockchain Now, to make this even more impressive, these models did this with ONE TOOL: Python. Not series of tools. Not MatLab. Not supercomputers. Now let me underscore what this means in the long run: your smartphone will be a math genius before too long. And a coding genius. And a linguistics genius. And... and... Third, fourth, and fifth order consequences of this one technology alone are impossible to overstate, and this technology will only get better. You know that scene where Tony Stark is figuring out time travel in his kitchen? Yeah, that's the level of AI math we're talking about in just one or two more generations. If warp drive is possible, these machines will help us figure it out.

I bet Anthropic is sitting on 3.5 Opus until GPT-5 drops. It's like a Mexican standoff between Anthropic, OpenAI, and Google

Does it strike anyone else as weird or coincidental that UFO activity is spiking right around the moment we solved AI reasoning?

o3 really really really wants to build stuff. It's like "bro can you PLEASE just let me code this up for you already???" It's like an over-eager employee. I'm not complaining.

This whole "no moat" situation in AI is fascinating, and it's probably the best thing that could have happened for humanity. Here's the deal: typically in tech, companies build these incredible fortresses around their innovations - patents, trade secrets, specialized hardware, you name it. It's how they keep competitors at bay and maintain their market position. But AI? It's completely different. Every time OpenAI drops something groundbreaking, within months you've got DeepSeek in China, Google, Microsoft, and a dozen other players replicating their results. The secret sauce isn't so secret anymore. There's no special hardware you need, no magical dataset that only one company has access to. The algorithms? They're largely based on public research. The data? It's mostly from the internet, which belongs to everyone. And this is incredible for consumers. Think about it - when you've got this level of competition, with everyone able to replicate everyone else's advances, what happens? Prices plummet. Innovation accelerates. No single company can get away with charging premium prices or restricting access because someone else will just come along and offer it cheaper or even free. The only real advantage anyone has in this space is speed - being first to market. But even that advantage is shrinking. What used to be a 12-24 month lead for OpenAI has shrunk to just a couple of months in some cases. Sure, they've got that first-mover advantage with ChatGPT being one of the most visited websites globally, but that's more about marketing and brand recognition than technological superiority. This is exactly how transformative technologies should work. Look at historical parallels - the printing press, radio, the internet. Once these technologies were understood, they spread everywhere because they were fundamentally democratic technologies. AI is following the same pattern, and it's beautiful because it means no single entity - whether that's a corporation or a government - can monopolize this power. We're watching the democratization of intelligence in real-time. And sure, some companies might not love this situation, but for humanity as a whole? This is exactly what we needed. When everyone has access to these tools, when anyone can innovate and build on top of them, that's when we'll see the real revolution begin. No gatekeepers, no artificial scarcity, just pure technological progress driven by global competition and collaboration.

America just won geopolitics for the next 50 years with Project Stargate. Let me explain why. This is where things get interesting - actually, scratch that, this is where things get revolutionary. We're about to witness something that happens maybe once in a generation: the full might of American innovation, capital, and willpower focused like a laser on a single, world-changing goal. Think about what the Manhattan Project gave us, or what NASA achieved when Kennedy threw down that moonshot challenge. When America decides something matters and backs it with this kind of money? It happens. Period. End of story. And this time we're not just shooting for the moon - we're reaching for something that will fundamentally reshape human civilization as we know it. The target couldn't be clearer: Build and deploy AGI. Simple, elegant, and absolutely transformative. But here's what everyone's missing - this isn't just about building the most powerful AI system in history. This is about ensuring American technological dominance for the next half century. Those ten data centers we're building? They're not just hardware - they're the new Manhattan Project labs, the new Cape Canaveral, and they're being built right here in the good old USA. Now, $500 billion might sound like a lot of money, but let me tell you what it's really buying us: imagine a billion Einsteins working around the clock on every problem humanity has ever faced. One Einstein gave us relativity and fundamentally changed our understanding of the universe. What happens when you multiply that by a billion? Every single doctor gets unlimited cognitive resources for diagnosis and treatment. Every scientist gets an Einstein-level partner working on their research. Every economic challenge gets solved by superintelligent analysis. We're not just removing the intelligence bottleneck - we're shattering it completely. But here's the real kicker, and pay attention because this is important: World War 3 just became obsolete before it even started. The first world wars were won with steel and manufacturing. The Cold War was all about nukes and economics. But this war? It's already over, and America won it with silicon and algorithms. China's entire IT infrastructure is about to look like a pocket calculator compared to what we're building in Texas. And speaking of Texas - there's a reason we're starting there. Everything's bigger in Texas, including our ambitions for the future. This isn't just acceleration - this is America taking the fast lane into a future where we write the rules. And let me tell you, it's going to be one hell of a ride.

Google and Microsoft just dropped twin bombshells. Microsoft: Majorana 1 - a scalable commercial quantum computing chip Google: Co-Scientist - Deep Research on steroids specifically for science

No science fiction has ever contemplated the world that we're building; one with ubiquitous superintelligence. Star Wars imagined droids everywhere and normalized, but they weren't particularly bright or clever. The Culture series imagined hyperintelligence, but only in centralized clusters behind the scenes. Star Trek predicted that computers would just be passive Alexas forever. This is why I am suspicious of anyone who claims to predict how AI will turn out with a time horizon any larger than a year or two. No one predicted how things would be, and which direction they are going.

Long COVID fucking sucks. I probably shouldn't share this publicly but you know what fuck it. It's my life and it's a fucking real thing. Here's what it's like to have a relapse: I was fine for about 3 weeks. Sleeping okay, even mowing the yard, doing pullups. But then the stress got to a tipping point, and it's two weeks of hell. Severe insomnia, sleeping only 4-5 hours some nights. Fine, I've survived worse. But then the jitteriness, headaches, and gut trouble start. And you're stuck on the couch for a week straight, and any effort - cognitive, physical, or emotional - is a full setback. Every time you think you're out of the woods, that you're pacing yourself and being careful, and watching your exertion like a hawk, it doesn't matter. "Recovery is nonlinear" They tell you that over and over and over again. But that's the reality. I mean, hell, I had three weeks of near normal life. That's a godsend. I haven't been hiking in nearly a year. So, not entirely normal for me. But I wasn't counting my steps and watching my heart rate like a diabetic watches their glucose. It started when my dog needed emergency surgery. That was all night at the emergency vet and $7000 I can't really afford. That pushed me over the edge and it was just too much. Ultimately, that stress is going to set my recovery back by like a month. Maybe more. At least this kinda thing is getting more publicity now. I've seen news articles talking about the hundreds of thousands or even millions of children struggling with it too. Anyone responsible for fucking Wuhan needs to be in prison for the rest of their lives and then some. Anyone doing gain of function research needs to be arrested and shut down. Forever. Millions of lives are fucking ruined. I hate everything right now. Fuck this. Fuck everything.

I hear people constantly claiming their jobs are secure - whether they're welders, farmers, or skilled tradespeople. Let me break down the financial reality that's driving the robotics revolution. We're looking at a $30 trillion annual market in skilled and unskilled labor alone. Any industrialist worth their salt is analyzing this opportunity with intense interest. Let's conduct a conservative thought experiment: if automation captures merely 10% of this market, that's a $3 trillion opportunity. Now, factoring in that robotic solutions will likely operate at one-tenth the cost of human labor, we're still examining a $300 billion addressable market. And that's just scratching the surface of manual labor - we haven't even touched the potential disruption in specialized fields like medicine and surgery. When you realize this $300 billion represents a mere 1% of the total labor market value, the economic imperative becomes crystal clear. The financial incentives to accelerate robotics development are astronomical. The mathematics of market dynamics don't lie - those dismissing this transformation are fundamentally misunderstanding the scale of economic forces at play. So, Mr. Plumber, Mr. Electrician, Mr. Carpenter, no, your job ain't safe. And I'm not saying this to be mean, I'm saying this to give you a little reality check so when the writing on the wall looks more like a billboard, you won't be shocked. I'm won't say "I told you so" because it's gonna be glorious for all of us, when the day comes that you set down your wrenches and pliers for good, I'll be there to welcome you to the afterlife - life after business and labor get a divorce. Breathe the free air, brother.

You don't actually want a job. You want money in the bank. You want a sense of mastery and purpose. You want autonomy and agency. Work will not set you free. The revolution is coming.

"But we still don't have AGI" says the scientifically and data illiterate ape

I sometimes forget that not everyone is as immersed in this stuff as I am, so allow me to explain what I mean when I say this is all over in 20 years anyways. This is not an arbitrary measure. 2045 is a very specific year. Specifically the attached graph. Many things become possible as compute scales exponentially. But why 2045? 2045 is the date by which a typical computer will have as much computational power as the entire human race combined. At this point, all bets are off. Any remaining scientific or intellectual problem is basically solved. Yes, some practical experiments will still take some time (building a ring habitat in orbit for instance). But at the same time, anything that can be simulated or modeled is 100% solved at that point. That means all disease are gone. That means that natural evolution becomes meaningless. So whenever someone says "Oh, X is at least decades away" you know that they haven't done the reading. They haven't even looked at the homework. No. You cannot predict beyond 2045, full stop. Everything we think is a problem today is over and done with before 2045. Yes, I know I expressed some skepticism before (see my previous video "Singularity Canceled") but I tried that idea on for size, and it simply didn't fit. The idea of one S-curve feeding into the next is looking more and more legitimate. Computational scaling laws have held out for an astounding 120 years. From mechanical to electromechanical, to vacuum, to transistors. Consider the rise of quantum, photonic, and thermodynamic computing. We have not one, but three rising new paradigms of computation. Okay, so Moore's law has been going for 12 decades. We just need two more, and it's all said and done.

I recently announced I was quitting AI but didn't explain why clearly. My reasoning centers on attractor states: extinction, collapse, cyberpunk dystopia, status quo, solarpunk utopia, and FALSC. My recent conversations with AI researchers confirmed we've departed the extinction and collapse trajectories. The AI safety movement succeeded - we taught, inspired, connected key people, and shifted the trajectory. I believe we're now clearly moving toward the cyberpunk dystopia attractor instead. Economists dismiss AI impact and double down on neoliberalism while Moore's Law shows accelerating AI compute growth across 12 decades. When AI and robots become the primary means of production, capitalists will own all production and the social contract collapses. That's why I'm focusing on post-labor economics and preventing cyberpunk dystopia.

Sorry Anthropic. I stood by you a long time. You were doing something different. But then you got safetypilled. Focus on product. You treat your users like children. UX is king. Intelligence is king. Your market share is falling, and deservedly so. This is your own fault.

Metacognition. That's what o3 and r1 are demonstrating. That's why everyone is freaking out. I don't mean this in the "functional sentience" perspective (though you could read it this way). I mean this in the "can solve problems ultra effectively" kind of way. OpenAI and DeepSeek figured it out at the same time. Though, to be fair, DeepSeek might have just stolen it.

If you want to cure aging and disease? Go all in on AI. If you want to solve climate change? Go all in on AI. If you want to end poverty and hunger? Go all in on AI. If you want to stop wars and genocide? Go all in on AI. If you want to invent nuclear fusion and explore the universe? Go all in on AI. Guys, this is not just another "general purpose technology" this is the the KEYSTONE technology to humanity's future.

Thermodynamic computing is here There is a new computing paradigm emerging from the noise, and its arrival may be as significant as the dawn of deep learning or the advent of cloud virtualization. A new company, Extropic, has just launched its first thermodynamic computer, a device they call a TSU, or Thermal Sampling Unit. While the web is already filling with deep technical dives, what’s more important for most of us is building a clear intuition for what this technology is, how it’s fundamentally different from anything that’s come before, and why it’s generating so much excitement. This isn’t just another chip; it’s a new way to think about computation itself. Seeing is Believing: Solving Puzzles in One Shot To understand what a TSU does, let’s look at two classic, notoriously difficult computer science problems: Sudoku and the Eight Queens problem. When you or I solve a Sudoku, we use a process of sequential logic, guess-and-check, and backtracking. We make an assumption, follow its logical conclusion, and if we hit a dead end, we erase and try again. A classical computer does the same, just much faster. A TSU, however, approaches this in a completely different way. Using a TSU simulator, one can “program” the problem by first clamping the known values—the clues already on the board. Then, you program in the constraints: no duplicate numbers in any row, column, or 3x3 square. With the problem thus defined, the TSU doesn’t “search” for a solution; it anneals one. In a single computational step, the solution simply emerges, backfilling all the empty squares correctly. The same principle applies to the Eight Queens problem, a challenge to place eight queens on a chessboard so that none can attack any other. This is a complex combinatorial problem with 92 distinct solutions. A classical computer would have to iteratively search for these. A TSU, by contrast, can be programmed with the constraints (the “anti-affinity” between queens on the same row, column, or diagonal) and then set to sample the “solution space.” In this context, a valid solution is one with a “problem energy” of zero. The TSU’s physical nature allows it to naturally find these zero-energy states. A simulation of this process shows the TSU discovering all 92 unique solutions, demonstrating its ability to not just find an answer, but to explore the entire landscape of all correct answers. This is a fundamentally new approach, one that bypasses the brute-force, iterative methods we’ve relied on for decades. The Physics of Computation: Using Noise, Not Fighting It This new power comes from a radical design philosophy. For the last 70 years, computing has been about one thing: order. We build chips that are deterministic, logical, and precise. The great enemy has always been noise, heat, and randomness. We spend billions on cooling and error correction to eliminate these very things. Quantum computing, in many ways, is the ultimate expression of this, requiring temperatures near absolute zero to eliminate all thermal noise and achieve quantum coherence. Thermodynamic computing is the polar opposite. It doesn’t fight the noise; it uses it. The TSU is built on the understanding that the natural, stochastic noise from “leaky” transistors—the very randomness we’ve tried to engineer out of existence—is itself a powerful computational resource. Think of it this way: a GPU, which is central to today’s AI, has to simulate noise. When a generative AI model creates a new image or sentence, it’s using complex algorithms to fake randomness. The TSU doesn’t need to fake it; it harnesses the actual physical randomness of thermodynamics. It is a piece of hardware that directly computes with probability. This makes it a hybrid, sitting somewhere between a purely analog computer (which might use light or sound waves to compute) and a digital GPU. It’s a physical device that leverages the laws of physics itself to find solutions, rather than just using logic gates to simulate them. From a Lost Hiker to a Million Bouncy Balls Perhaps the best way to build intuition is with a metaphor. Imagine that solving a complex optimization problem is like trying to find the lowest point of altitude in a 100-square-mile mountainous landscape. Classical computing, using an algorithm like gradient descent, is like being a single hiker dropped into this landscape at night. You have no map or satellite view. All you have is an altimeter and the sensation of the slope under your feet. You can only take one step at a time, always walking downhill, hoping you don’t get stuck in a small local valley when the true, lowest canyon is miles away. Thermodynamic computing is a completely different approach. It’s like having a million bouncy balls and a helicopter. You drop all million balls simultaneously across the entire 100-square-mile landscape. Then, you “turn on an earthquake,” shaking the entire system. The balls bounce and jostle, but as the shaking (the “annealing”) subsides, where do they all end up? They naturally settle into the lowest points. The balls that collect in the deepest valley represent the optimal solution. The TSU is, in essence, a physical device for dropping those million balls at once and letting the laws of thermodynamics find the lowest “energy” state for you, all at the same time. Beyond Puzzles: The Real-World Impact This is far more than just a clever way to solve brain teasers. This ability to instantly find the lowest energy state for a complex, constrained system has staggering real-world applications. One of the most immediate is protein folding. Companies like Google’s DeepMind have made incredible progress with AI like AlphaFold, which predicts protein structures. But this is still a predictive model trained on existing data. A TSU could potentially solve the folding problem directly, treating the protein as a system of atomic affinities and repulsions and finding its most stable, lowest-energy configuration almost instantaneously. This could revolutionize drug discovery and materials science. An even more profound possibility lies in nuclear fusion. One of the greatest engineering challenges in history is controlling the superheated plasma within a tokamak reactor. This requires shaping unimaginably complex magnetic containment fields in real-time to prevent the plasma from touching the reactor walls. This is a real-time optimization problem so complex it’s currently beyond our capabilities. A TSU, however, could be fast enough. Its ability to compute with electricity itself, rather than abstracting the problem through layers of software, might allow it to update the magnetic fields fast enough to stabilize the fusion reaction. One could even imagine a future where thermodynamic computing elements are built directly into the tokamak’s walls, allowing the reactor to physically and intelligently react to the plasma’s state in real time. A ‘GPT-2 Moment’ for a New Era It’s easy to become numb to hype, but what we are witnessing with the TSU feels different. This is what you might call a “GPT-2 moment.” For those who were there, GPT-2 was the first generative AI model that wasn’t just a toy; it was the first time you could play with it at home and see the spark of true generative intelligence. It was the precursor that pointed directly to the GPT-3 and ChatGPT revolution that has since changed the world. This TSU has that same feel. It’s the “SDK” for a new computing paradigm. This technology is as different from classical computing as quantum computing is, but with a critical difference: a team of 15 built this in two years, and it runs at room temperature on your desk. Quantum computing has seen decades of work and billions in funding, and it still hasn’t produced a commercially viable, scalable machine. The TSU is here now. Based on a two-decade-long career at the cutting edge of technology—from seeing the obvious future of virtualization in 2007 to an early conviction in deep learning and GPT—this has all the same hallmarks of a fundamental, world-changing shift. We are not just building faster calculators; we are learning to compute with the universe itself. Pay close attention to this. This is the next big thing.

What many people are not yet getting is that advanced AI nullifies all jobs. The marginal value add of any human input eventually drops negative. We saw this with chess, for instance. Originally, human/AI hybrid teams were superior. But now, humans just add noise. Right now, we're still in the human/AI hybrid phase of business. But eventually, we will just be dead weight in all roles.

It's an absolute privilege to be alive right now. I mean, it's a privilege to be alive at all, but in particular this moment. All human concerns will seem trivial soon.

No. The marginal utility of human labor has been dropping for decades, along with the demand for human labor. People will still have "work" as in "effortful things they find meaningful" sure. But "jobs" as in "a wage labor arrangement with an employer" is going away for the vast majority of people. We need to stop deluding ourselves that the current employer/employee relationship is either durable or desirable. It is neither.

Within a few months, OpenAI is going to dominate all leaderboards in math and coding. Other frontier AI labs will follow suit. It's the knee of the curve. All stations, green for liftoff.

I've had my criticism of Sam Altman but now that he's saying AGI in 2025, I'm just like "Well yeah, duh." Just look at the data. Look at the scaling laws. Look at the trends. Unless something goes catastrophically wrong, it's inevitable. It will take longer to integrate AGI into the economy and society than it will take to achieve AGI. That most people are still unaware of what's happening is insane. If you see this tweet, you're at the cutting edge. That's another reason I'm so freaking excited to be working with @JuliaEMcCoy on our Leaders of AI podcast and our book #TheGreatDecoupling. The future is looking brighter than ever, even if we have some problems to solve along the way.

o3-pro encouraging 4o

OpenAI is not a serious company any longer.

Tens of thousands of general purpose humanoid robots will be shipped by end of 2025. Hundreds of thousands in 2026. Millions in 2027. By the end of 2026, those robots will have an effective IQ of greater than 160. They will be universal polymaths, each one individually more competent than the top 100 humans across any domain. Any nation not committing to full acceleration in this paradigm is committing sovereign suicide. United States of Acceleration is go for liftoff.

Me refreshing chatgpt all day to see when I get GPT-5.

Yeah, OpenAI has lost the Mandate for sure

Anthropic has taught Claude to use a computer directly. I've been saying for the past two years "If you can do your job exclusively in front of a computer, you're the first to go." Learn to work with humans if you want to be in the job market for any length of time. People skills will continue to be the most valuable skills in business, and eventually, the primary skillset of the Meaning Economy. Tell a friend.

A postdoc (PhD) friend of mine texted after testing o1-preview (Strawberry) against a very tough CFD problem (computational fluid dynamics) and it almost got it right the first time. Got it right with some feedback. ChatGPT was already useful to him, and he used it all day to help with coding. I suspect that o1-preview and then o1-(full?) or whatever comes out in October/December will be even more useful. This will accelerate science across the board.

Some words of wisdom for Sam: Never ever use the term "compute budgets" again. That's like saying "us tech elites will give you an allowance of our toys, peasant." Notice how Sam is always hedging with words like: "Ensuring that the benefits of AGI are broadly distributed is critical." Or, you know, you could just distribute the AGI? Software is free to infinitely copy. This post reads like doublespeak.

Okay real talk for a second. China duplicated OpenAI's o1-preview in 2 months. I'm talking about DeepSeek. The math and algorithms are not a moat for companies... Or for nations. This Cold War is gonna be very different. So how do we (Western powers with liberal democratic values) win? We've got to win on economic fundamentals. That means more chips, more data centers, more energy. Faster Internet, more saturation and connection. This also means we need enabling infrastructure and legislation. We need to be moving at the speed of AI, not the speed of humans. We need to create a new form of commerce and innovation, one based on AI agents as soon as possible. I call this A2A or "agent to agent" This is the way

Sam Altman: "We're gonna deploy infinitely scalable beyond-PhD level SuperAgents" Also Sam Altman: "We're not deploying AGI" Eventually you're just playing word games.

"AI is hitting a wall!" No, it's just smarter than the average user now.

Let me lay this out fresh, because after years of building cognitive architectures, I've learned something fascinating about intelligence versus agency. Everyone's hyping up AI agents like they're the next breakthrough, but they're missing a crucial piece of the puzzle that took me way too long to figure out. Here's what threw me for a loop: raw intelligence isn't the same as knowing what to do with it. Think about it - chatbots were relatively straightforward because they're just waiting for humans to start the conversation. Give them a prompt, they'll dazzle you with their smarts. I call this the "genius in a jar" phase. But true agency? That's where things get spicy. The real challenge hit me on day one of building cognitive architectures. You're sitting there with this incredibly powerful system that can think about literally anything, and suddenly you realize - wait, how does it decide what to think about? It's like having infinite processing power but no executive function. The human brain handles this naturally, but programming it? That's a whole different ballgame. You'd think solving the "what to do" problem would be straightforward with enough computing power, right? Wrong. I had to dive deep into ethics and philosophy just to create a basic decision framework. Utilitarianism, teleology, deontology - throw all that in and you've barely scratched the surface. Because then you need to build out the entire model of what the agent can and can't do, what it should and shouldn't do, and most importantly, why. That's the "agent model" layer - it's gotta know "what I am?" and "how do I work?" before it can start even using its digital hands. I call this the "constraints, capabilities, and context" or "Three C's" of the agent model. And just when you think you're getting somewhere, you hit the broader context problem. Your agent needs to understand its environment, available tools, specific tasks, access levels - and we haven't even touched goal tracking or task prioritization yet. It's like trying to build JARVIS from scratch and realizing why Turing never cracked this part of the puzzle. Resource utilization, keeping track of what you need and have, and how close you are to solving the actual problem. LLMs can do each of these steps no problem, but coordinating it all? A bit harder. That's what makes today's LLMs so interesting - they're still narrow AI in their operation, but they're packing some serious general intelligence features under the hood. World models, reasoning, planning, problem-solving - it's all baked in. But true agency? That's still the holy grail we're chasing. And it's not necessarily just the model, it's the rest of the "body" so to speak. Think about it, if your brain was just floating in a jar somewhere, no eyes, ears, or hands, you couldn't really do anything either? To make matters a bit more confusing, AI agents exist in cyberspace, where what they 'see' and 'hear' is just too different from our meatspace context, and their 'hands' are API calls. Look, here's the punchline that most people haven't wrapped their heads around yet: We cracked this problem. Those years of research into cognitive architectures and agent frameworks weren't just academic exercises - we were building the foundation for something exponentially bigger than anything we've seen before. The real kicker? Once you've got one working agent system, you don't just have one - you've got the potential for billions. The same agent, perfectly replicated, operating at scale across every digital system on Earth. Copy, paste, repeat. We're not just talking about replacing a few mid-level engineers here and there. We're talking about a tsunami of artificial executive function that's about to reshape every industry, every workflow, every digital interaction. The people tweeting about 2025 aren't being optimistic - if anything, they might be underestimating just how fast this is going to move once it starts. The dominoes are already falling, and trust me, you're going to want to be ready when they hit.

So, fun fact, to the model, the entire chat log is just one blob of text. It requires quite a bit of training (and often some stop tokens) to make the chatbot stop taking both sides. Essentially, what you do when training a chatbot is just train it on a ton of conversations (or RL-based conversations) so it really has no idea where "you" end and "it" begins. Most of that is just stop tokens baked into training. In other words, the chatbot is constantly predicting both sides of the conversation, it's just that us pesky humans halt it and inject our own tokens. What you've done here is make it avoid the stop tokens and use it's theory of mind to anticipate your next move (which it was doing already). Hope this helps. Source: been training these things since GPT-2.

The fact that DeepSeek mogged OpenAI into giving o3-mini out for free is interesting. Now that agents and operators are out, let me tell you a bit about what's going to happen next. First, people will need to learn to use these tools. But keep in mind that these are just the first iteration of these tools. New best practices will need to be established both on the user side as well as the server side. But now that OpenAI has demonstrated a thing, all the competitors will be scrambling to the duplicate it. Within 6 months there will be several commercial alternatives, as well as at least one or two open source alternatives reaching parity. But this doesn't even begin to underscore the compounding returns we're about to see. Productive will start to climb for everyone, businesses will adopt, scientists and researchers will adopt, and it will just keep going. Remember, everything we've seen up until this point has just been chatbots. We're seeing the first generation of semi autonomous AI being rolled out en masse right now. It seems I have a few videos to catch up on...

Umm... GPT-5 just corrected itself midstream. I've never seen an AI do this. That kind of self-monitoring and error correction is, until this point, a distinctly human trait. I don't think it was an affectation, either. This implies a level of cognitive control that may be emergent in AI. Has anyone studied this? (note: this was vanilla GPT-5 without thinking)

I think my biggest guilty pleasure right now is the schadenfreude of mediocre people realizing just how mediocre they are in the face of AI. AI isn't just a mirror. It's a big self-awareness stick, and it's coming right down on them.

Beauty standards have gone too far