Really enjoyed reading the Microsoft MAI-Thinking-1 "Building a Hill Climbing Machine" paper. Amazing they publicly released all the info needed to train a frontier model, down to hparams.
I also thought this was pretty telling:
- pre-training: 30 trillion tokens
- mid-training (SFT on STEM/math/code data): 3.55 trillion tokens
- RL post-training: 150 billion tokens.
Looks like @ylecun was right all along with the cake analogy.
Obviously I still think something like RL (optimizing for long term goals) is fundamental to what we think of as intelligence. But it's not the volume of learning signal, it's the optimization on top of an already reasonable predictive model.

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development
"Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning."
Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing.
This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider.
That is not safety. Safety policies should be transparent, auditable, and user-visible.
On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

One shot procedural graphics handled perfectly by Claude Fable 5 + Replit, all created while sitting in a Tesla!

Concentration of power, capabilities and economic wealth is the biggest risk in AI. We need open science and open-source more than ever!

Did my (very not scientific) F-Zero test for Anthropic's new Fable model.
It didn't quite one shot below, I had to give a few more prompts like "give it more of a sense of speed." But still impressive!
Quote tweeting what GPT 5.5 made from 2 months ago.
Peter Yang: I've been doing the F-Zero test each time a new model comes out and GPT 5.5 and Codex is the only combo that built a working game so far.
Even made a bunch of other bots to race against. What a insane time to be building.

When you hear AI "safety" you should hear "censorship" and "control" instead.
All of us surveilled and spied by safeguards of loving grace.
Today it's intelligent Terms of Service control. You can't do AI research. Can't ask this question about your kid's biology homework.
Tomorrow it's refusal to help you with competing coding projects. A 100 page blacklist of questions.
Or this question means AI will search your computer stealthily and snitch on you to the cops because you posted an prohibited insult in a WhatsApp chat in the UK.
Open source must win out at both the model and the harness level.
That's because AI will become our interface to the world.
It will sit higher in the stack than the OS. It will collapse current SaaS layers, chat, communications, apps, app creation, into a single new kind of interface that doesn't exist yet.
It's got to be open. It's got to be a cypherpunk solution that makes privacy and security the number one priority.
If a closed source solution wins this layer, it's a disaster for the world. Especially if it's built by a single company with a single closed source model.
Why?
Because what we share with AI will be more intimate than anything we've ever shared with a machine. It will be our friend, our sounding board, our advisor. It will know our business ideas before we've told anyone. Our medical issues. Our financial picture. We'll talk about the fight we had with our partner. About feeling lost or depressed. Our kids will talk to it about problems at school, about bullying, about heartbreak, things they won't tell us.
It will know us more intimately than we know ourselves.
Right now the world runs on a surveillance economy. We traded free stuff for apps that peer deeply into our lives. If we replicate that model in the AI era, it's not just surveillance economy 2.0.
It's surveillance economy squared.
Social scoring. Automated evidence gathering. Legal conversations you thought were privileged showing up in court. Random people making $2 bucks an hour on the backend from God knows where reading the most intimate details of your life.
Every insecurity, every fear, every half-formed thought you whispered to your AI buddy at 2 AM, sitting in a database somewhere, searchable.
If we let closed source models dictate what we can and can't do it will only get worse and worse.
We've got to fight this future with every last breath.
If you can read this, you are the revolution.

just to state the obvious: think there's a collison course between those who believe research and science should be open and those who believe we are in an accelerating singularity curve.
I have many smart friends who have believed both for a while but seeing more and more their realization that these beliefs will be in conflict.
I for one believe that America and the west needs open and distributed access to research and computation and sharing of ideas at all times.

"It's also irresponsible," Michele Catasta, Replit's president and head of AI, said of "tokenmaxxing." https://bit.ly/3SvDAKQ

You have Claude Fable for only a few days. Here's how to make the most of it.
Introducing /improve: use your most capable model to audit your codebase and write plans for cheaper models to execute later.
Studies your code, figures out bugs, perf, tech debt, missing tests, what to build and writes plans any agent can run.

I talked to a founder of an AI startup generating about a 40% annual return on the cost of the GPUs he was using. I.e. he could make $400 in annual revenue for every $1000 worth of hardware he used.

Re @ClementDelangue @Dan_Jeffries1 Everyone, please join Project Tapestry
https://thealliance.ai/projects/tapestry

A week ago, I left my product job at Roblox to bet on myself.
This wasn’t an easy decision. After over a decade in product, walking away from a great job felt like giving up a core part of my identity.
I’m grateful for my 3 years at Roblox and have many fond memories shipping products with my talented team of engineers, designers, and PMs. Most of all, I loved building with our creator community and watching their eyes light up as we put our products in their hands.
At the same time, I was spending nights and weekends writing about AI, interviewing AI builders, and learning to build with agents myself.
What finally helped me decide were 5 principles I wrote down after a lot of reflection.
I made a very personal video walking through those principles. Maybe it can help you think through your next career move too.
📌 Watch here: https://www.youtube.com/watch?v=T4pvKvAE_SA

if you think fable is incremental you are not being ambitious enough with how you prompt it

Pedro is living at the edge more than any founder CEO I know. He inspired me to free my own Openclaw and you can see here how deep in the technical details he is (before selling Brex for $5.5b he was the first person to jailbreak the iPhone, as a 12 year old).
Y Combinator: Brex co-founder and CEO Pedro Franceschi believes most people still underestimate how much AI will change the way companies are built. AI isn't just another tool, it's a new foundation for building products, teams, and companies.
In this episode of @LightconePod, @pedroh96

Yann Lecun published the most heretical AI paper of the year.
He opens by arguing Magnus Carlsen isn't good at chess and only gets more unhinged from there.
The Turing Award winner and his co-authors dropped a paper demanding the AI industry abandon its biggest obsession, AGI.
Right now, everyone from Silicon Valley CEOs to politicians assumes AGI is the ultimate goal. A machine that can do everything a human can do.
LeCun argues that this entire concept is a biological illusion.
Humans do not possess "general" intelligence. We are highly specialized biological machines, tuned by evolution simply to survive in the physical world.
We only think our intelligence is "general" because we are completely blind to the millions of cognitive tasks we are incapable of comprehending.
Which brings us to the chess argument.
Magnus Carlsen is the greatest human chess player in history. But compared to a modern computer? He is fundamentally terrible.
Our belief that Carlsen is "good" at chess is pure human-centric bias. He isn't objectively good. He's just better than the rest of us, who are biologically awful at it.
LeCun says we need to stop building AI to mimic human generality.
Instead, he proposes a new North Star: SAI.
Superhuman Adaptable Intelligence.
Instead of trying to build a machine that mimics our flawed, biologically-limited brains, we need to embrace extreme specialization.
SAI is about the speed of adaptation.
It is an intelligence that can learn to exceed humans at any specific, economically important task.
More importantly, it is designed to fill the vast skill gaps where humans are fundamentally incapable.
Things like managing global energy grids in real-time. Or predicting complex molecular structures.
The entire AI industry is obsessed with building a digital reflection in our own image.
LeCun's paper is a brutal wake-up call.

I quickly became friends with Arcee's leadership and can't help but root for their humble approach to building the open ecosystem. No nonsense licenses, no projecting, just enabling broad access to efficient intelligence.
I'm happily supporting their research as an advisor.
Arcee.ai: We are thrilled to announce that @natolambert is joining Arcee as a Research Advisor.
Nathan’s work and thought leadership have been instrumental to the open model ecosystem, and his guidance comes at a critical time as open builders face growing pressure.
This is a major

Congratulations to Allen Liu, Assistant Professor of Computer Science, who has received the ACM Doctoral Dissertation Award!

🔥 Socket Firewall is now built into @Replit's AI-powered development experience.

It’s already blocking 8K malicious packages/day across builders on the platform, giving Replit users stronger protection by default the moment dependencies are introduced.
https://socket.dev/blog/socket-partners-with-replit-to-block-malicious-packages

thrilled to finally announce something I've been working on for a while:
@SocketSecurity is officially powering @Replit’s new Package Firewall!
By evaluating dependencies directly at the install path, we are protecting builders from hallucinated or malicious packages before they can execute. We're currently blocking 8,000+ bad packages a day across builders on Replit.
Ship fast, vibe safely. 🛡️
Read the full breakdown: https://socket.dev/blog/socket-partners-with-replit-to-block-malicious-packages

Introducing @PoeticHQ: a new AI system that executes complex multi-hour tasks with 99%+ accuracy and 10x fewer tokens than agents.
We raised $50M at $500M from Kleiner Perkins, Founders Fund, First Harmonic, and Genius Ventures to build AI that does complex work inside Fortune 500 companies without hallucination.
While code is too brittle, agents are too unpredictable. The work that runs the global economy - anti-money laundering, fraud investigations, underwriting - needs extreme accuracy.
So we built a new kind of software that pairs the flexibility of AI with the predictability of code.
When the world stays the same, Poetic runs fixed code: fast, cheap, identical every time. When the world changes, Poetic uses AI to regenerate its approach and find its way back to the objective.
In one year, we went from zero to an eight-figure run rate as a team of four.
Since then, we’ve scaled the team and executed the highest-stakes processes at AIG, SoFi, and Chime. At SoFi, a large US bank, Poetic reached 99%+ quality on fraud investigations in five weeks.

Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap: https://darioamodei.com/post/policy-on-the-ai-exponential

Re TLDR:
1. Declare AI too dangerous for ordinary competition so you propose a regulatory regime where only the largest incumbents can survive
2. Warn about labor displacement while selling the product to executives as a labor-displacement tool
3. Warn about state overreach while asking the state to license and gatekeep frontier models
4. Warn about corporate power while sketching a corporate-state cartel over compute, release, security, export controls, and deployment

interesting segment on bio in Dario's essay. specifically the part below. I'd be very curious how he imagines AI can be used for PD/PK modeling and tox.
I assume he means "ask claude for whether a molecule will be toxic" but the existing data on the web for drug tox is probably not sufficient to be good at this, and only reaching human med-chemist-expert level would be effectively random.
would be cool to see them pay to generate a massive new dataset 1000x bigger than anyone else and throw it into the training set though.

Peter is one of the nicest person in tech and I could not be happier seeing him take this plunge.
He is right, a lot of builders are currently facing this dilemma.
The largest companies are not setup well to create space for pure builders.
Peter Yang: A week ago, I left my product job at Roblox to bet on myself.
This wasn’t an easy decision. After over a decade in product, walking away from a great job felt like giving up a core part of my identity.
I’m grateful for my 3 years at Roblox and have many fond memories shipping

Give yourself permission to build.
The traditional career ladder pushes everyone to become a leader, but I just want to be a builder.
As you climb the ladder at most companies, you’re expected to step away from building and fill your time with product reviews, cross-functional alignment, managing up, and performance calibrations.
I know a lot of builders who spent their best years climbing the wrong ladder.
The good news is that this is finally changing. Companies are rewarding builders and ICs more than ever, and even managers are increasingly expected to do IC work too.
But becoming a good builder takes reps, and it’s hard to put in those reps when you’re in back-to-back meetings all day.
So if you’re a builder at heart, embrace it. You don’t have to give up what you’re good at to be a “leader.”
📌 Watch now: https://www.youtube.com/watch?v=T4pvKvAE_SA
Peter Yang: A week ago, I left my product job at Roblox to bet on myself.
This wasn’t an easy decision. After over a decade in product, walking away from a great job felt like giving up a core part of my identity.
I’m grateful for my 3 years at Roblox and have many fond memories shipping

Jordan-born AI billionaire Amjad Masad moved to the U.S. in 2012 at age 24 and founded vibe coding outfit Replit with his wife, Haya Odeh, in California four years later. In March, it was valued at $9 billion.
Read more about American immigrants leading in their industries: https://www.forbes.com/sites/giacomotognini/2026/06/09/america-doesnt-work-without-immigrants-heres-why/?utm_source=ForbesMainTwitter&utm_medium=social&utm_campaign=ForbesMainTwitter
#Forbes250
Photo: Robert Severi for Forbes

The videos from the “Frontiers of Embodied AI” meetup at ETHZ from a few weeks back are now available.
Speakers: Jitendra Malik, Vladlen Koltun, Yann LeCun, and Shuran Song
Hosted by Marc Pollefeys
YouTube playlist: https://youtube.com/playlist?list=PLfjJj_IgRo7DWoamlTwlK7-4bZhNNNFgM&si=Z3_cfNB6Ijg6_PWN

After spending more time down this rabbit hole lately, it seems like @NousResearch Hermes + GBrain (+ Obsidian + GitHub) is the most optimal path
I’ve been setting up an Hermes agent recently and I am wildly impressed by how good it is. With building a strong foundation that is portable + scalable + lightweight, it’s become clear where this direction is heading and I don’t feel pressured by model lock-in
This is 100% the future of agentic workflows
Connor King: some f/u thoughts:
the active problem is that as agents start taking real actions (eg agentic robinhood — trading on your behalf, managing accounts, paying), the agent's memory of you basically *is the agent. risk tolerance, prior decisions, what worked, what didnt work last

This is a blatant misrepresentation of why SF Pretrial non-profit is losing its contract with San Francisco. They lost their contract because they got caught stealing their own employees retirement contributions. It was in the news and everything. 🤷‍♂️ https://www.nbcbayarea.com/news/local/san-francisco-pretrial-diversion-project-retirement/3812447/
NowThis Impact: SF is replacing 105 community workers with just 27 armed probation officers to supervise people who haven’t been convicted of anything.

🚨BIG NEWS🚨 The wait is officially OVER! 🚀 You can now push posts directly from @PomelliByGoogle to Instagram!
Yes, really. Reels, Stories, and static posts - straight from your Campaign Creatives and Photoshoots to your feed.
Check out how easy it is! 👇

BREAKING: In a stunning moment, Donald Trump's former economic advisor just admitted on Fox News that inflation is being driven nearly exclusively by bad decisions that Donald Trump has made. Wow.

The Model Lab vs Agent Lab distinction is one of the clearest frameworks I've seen for understanding where AI value actually lives right now.
TL;DR from @latentspacepod:
• Model Labs compete on capability (tokens, benchmarks, reasoning)
swyx: New @latentspacepod Essay:
why Agent Labs are clearly emerging in 2025 as a complement to Model Labs' all becoming AI Cloud platforms.
https://latent.space/p/agent-labs

Everyone Operating At The Frontier
Satya Nadella, Chairman & CEO, Microsoft, interviewed by @saranormous & @eladgil (No Priors) and @swyx (Latent Space)
Crossover special at Microsoft Build 2026.
Summary: Satya reframes Microsoft's AI strategy as an ecosystem play rather than a single model or platform, where the win is any company being able to point to AI it created and operate at the frontier with its own intelligence. Scaling laws held and intelligence still tracks the log of compute, but the value lives in deployment, where private evals become a company's biggest IP and accumulated agent traces start to look like assets on the balance sheet. Take it seriously and SaaS gets unbundled and rebundled, engineering collapses toward generalists who manage agents, and the industry has to earn community permission for the buildout by delivering benefits people can actually see.
1. Ecosystem Over Model. A platform earns its place by how much value other companies build on top of it. Satya wants any company, AI-native or traditional enterprise, to participate as a first-class participant that can point to AI it created, still using other people's models but owning a recipe of its own. He calls this the only tagline that matters for the conference: can everybody operate at the frontier with their own frontier intelligence. Without that, he says, there is no reason to hold a developer conference; you would just "worship at the altar of one model."
2. The Broken IDE. Coding agents worked so well that Microsoft now has to rebuild the IDE around them. When a developer runs a hundred agent sessions at once, the cognitive load lands back on the human and chat as the only artifact stops working, which is why the new interface needs a canvas. Even a fully agentic world still needs UI, because someone has to inspect what the agents did and decide. The lesson generalizes: every workflow handed to long-running agents will need a new surface for the human to supervise it.
3. The Harness Is The Product. The unit that matters is the harness that loops across models, data, and tools. Microsoft runs the same open GitHub harness across GitHub Copilot, security copilot, and science discovery, with progressive disclosure of tools to stay token-efficient and heavy context prep where "the magic is." The harness stays open: bring your own models, tools, and context, or swap in a Llama harness. Nadella points to M-dash finding vulnerabilities the incumbent scanner missed as proof that a multimodal harness can win in the real world.
4. Private Evals As IP. The single most valuable thing a company can own is a private eval. His acid test for control: take your private eval, run it on model A, then switch to model B; if you can still climb, you are in control, and if you cannot, you are not. Because frontier models learn from a few samples rather than mountains of data, the defensible asset is the eval you never leak. This is why Nadella reframes Microsoft's third act from operating systems to cloud to an evals-and-harness company.
5. Agents On The Balance Sheet. The traces between a company's humans and its agents become a trainable asset that belongs on the balance sheet. Human capital never made it onto the balance sheet because tacit knowledge could not be captured, but agent traces collected over time can train a "company veteran" agent that encodes how that specific enterprise creates value. As token capital and human capital both rise, the question becomes how to compound the two. Elad Gil's quip lands the point: the SEC will need accounting standards for token expertise.
6. Unbundle And Rebundle. SaaS gets taken apart and put back together, with the data model and business logic surviving the teardown. A general ledger should stay a general ledger, and a Power BI semantic model is hard-won business logic worth feeding to agents, so the work is repackaging these into new bundles and business models. Work IQ exposes what Nadella calls the most important database in a company, the M365 data that was only ever captive to email and Office apps. Now an agent can read a week of design-meeting transcripts tied to a GitHub repo and come back with a plan to change the code base, something M365 was never built to do.
7. Outcome Pricing's Catch. Per-user pricing is an artifact of buyers needing budget certainty, and it survives even as consumption pricing arrives underneath it. Subscriptions bundle some usage into per-user stacks, then consumption metering sits below, which is exactly the adjustment GitHub made after agent intensity blew past what per-seat assumed. Outcome-based pricing sounds appealing until a customer actually has an outcome and realizes they are giving away a royalty. As Nadella puts it, most people love outcomes until they have one, then they ask to go back to per-user and consumption pricing.
8. The Buy-Or-Build Test. Whether to build software or buy it reduces to a quantifiable rule: acquire it when the marginal cost of building and maintaining it yourself is higher. Maintenance is the part teams forget, because security holes that AI now finds faster also have to be fixed faster, and every fix burns tokens that someone has to own. Satya expects the current agent euphoria, where teams rebuild everything internally, to cool after one full budget cycle. The vendors that last will be the flexible ones; he sees very little tolerance ahead for any vendor that stays rigid.
9. Generalists Win. The biggest returns go to generalists whose scope just grew. LinkedIn restructured into a "full stack builder" discipline that combines design, product, and front-end while keeping each person's original edge, giving people bigger scope instead of one narrow role. Building an app now sits in the same sentence as writing a Word doc or a spreadsheet, so generalist skills suddenly carry, in Satya's words, "a higher leverage." Specialists still exist, and infrastructure science, like building the RL environment where a reward can be learned, becomes one of the hardest and most valuable roles.
10. Meta-Work. The biggest move is to make your work meta: build the agentic system that does the work instead of doing the work. Satya's example is the team running Azure's physical fiber network, who decided their job was not Azure networking but building the agentic system that does Azure networking, complete with a named agent called Miles. That team started asking for tokens instead of headcount to scale their operation. Kevin Scott's line frames why it matters: making hard things easier is one kind of progress, but true ambition is making the impossible possible, and that needs a new conceptual model of what work even is.
11. Earned Permission. The industry only gets to keep building data centers if communities feel the benefits in real ways. Satya argues the buildout has to lower energy prices through a better long-term grid, replenish water through closed-loop systems, and show up as jobs and tax base, with the burden on the industry to earn that through hard work. His read on the politics is blunt: the world will be skeptical of any tech company that says "trust us, the future will be glorious," so you have to deliver tangible benefits people can see in the next 12 to 18 months. Using a lot of energy while creating a lot of value for society has historically been a good story, and he is betting a token economy that drives productivity and broad participation lands on the right side of it.
12. A New University. The next great startup may be a new university. Satya thinks the way we educate, credential, and value those credentials has to change completely now that the means of learning and staying current have shifted so fast. Learning concepts still matters, and he points approvingly to a Stanford AI class drilling students on when to apply softmax rather than just asking a model to fix a training run. The opening he sees is for someone to build a new way of teaching that takes a person through a curriculum and out the other side into real economic opportunity, something that felt impossible for a long time.