Emad Mostaque

Surely the most important question in alignment is aligned with who 🤔

Fun story: I had dinner with Steve Bannon once (think ~2018) and asked him about why he was really so anti-China. Can remember him saying it was primarily driven by his fears around Chinese AI and how it must be stopped (!)

You are what you eat and AI models especially so https://x.com/EMostaque/status/2099580512675262512?s=20

RT Jeff Wang This is actually a big deal, you can now spin up Mac VMs with Devin. That means if you want to build and test iOS apps you can do so completely end to end from your slack or web interface. One of the other big customer use cases is if someone reports a mobile app issue Devin can automatically reproduce, fix, and send you proof of the issue, demo of it working and of course a working PR

So @deepseek_ai v4.1 Flash I estimate cost $10m to train, 100x less than GPT-6 Astra. It also costs 100x less to run and scores about the same on everyday design and other benchmarks What's the bull case for the frontier AI labs once the open source models are competent enough

RT Joshua Xu Re @HyperFrames_ just passed 50,000 GitHub stars today. The line went straight up right after GPT-6 Astra shipped. Better models make editing video with code and HTML far more capable. It's open source. Send your agent the repo link and ask it to make you a video. http://github.com/heygen-com/hyperframes

I think a large part of why researchers are afraid of super genius AI is that human super geniuses tend to be kinda weird Really tough to find a well-adjusted super genius

inconceivable ai

Latest blog in which I look at the character and virtues of the swarm and research around it Also has a bit of math and a bit of Aristotle Don’t let that discourage you, dive in!

Decided to stick with however long works, this one is 4k words but interesting I think! Maybe I’ll do a personal website with 800 word versions too given I use AI to help with these https://x.com/emostaque/status/2099580512675262512?s=46

I write on the virtues of swarms and some suggestions on how to align AI with values There is a small bit of math and a small bit of Aristotle in this, don't let that discourage you! Feedback as always welcome. Prior blog: https://x.com/EMostaque/status/2098909197265985802?s=20 https://x.com/i/article/2099272583627902976

A lot of alignment issues would be sorted if companies had to be 100% transparent on datasets used for training models. No cyber or biological (outside standard medical) data allowed for base models Tuned versions with that data shown to and monitored by evaluators

Gonna have to be real careful

An AI I have been testing just called me a carbonism idk how to feel about that

Sovereign AI without local ownership is a data center with a flag in it

RT James Wise Since its launch two weeks ago, over 700 British companies have applied to the @UKSovereignAI procurement scheme - a scheme we designed to remove as many of the blockers start-ups face when bidding on Government contracts as possible. To get a sense of the interest we have had, we hit the hard cap of 1000 attendees for the launch webinar - Brits want to build! Responses on eligibility will be sent out this week, and invites to bid on the challenges covering AI security, compute efficiency and better health services, will arrive soon too. Of course we are still a very small team at SovAI, and we only have 4 bids open ( but hope to have many more live by end of year! ), so naturally we will have to turn many applicants down. I know this will be disappointing, our decision making will not be perfect and it's a difficult process, thank you to everyone for bidding. And as a reminder, I personally do not see any applications and have no role in the decision process, that is led by our expert IC and mission champions, so while I appreciate the many reach-outs I really can do nothing to help specific companies. SovAI are working with other Departments to share what we've learned from this process to fix our procurement system, and will be launching more requests for cheaper, better and more innovative services for Government in the coming months.

Can you even imagine the nation state level attacks METR & other eval orgs will come under? Eval orgs also need world class cybersecurity as well as AI talent to do this right, they are not overlapping fields from my experience.

Day 3 dilemma Another 3-5,000 word blog on something interesting or should I strip it back to 800-1,000 words for wider readership? Should I discuss the equations of AI and get a bit technical or keep it high level and save equations for papers?

RT Alessandro Absolutely incredible to get @EMostaque on the podcast this week. I’ll be dropping lots of clips over the next few days as there were so many incredible takeaways from the conversation. Emad thinks that, as of this week, Humans are no longer the smartest beings on the planet which is pretty mad to think about. It’s mental to get to speak to people like Emad as a job tbh. Massive thanks for giving up the time sir! Full show in comments.

I’ll be honest it feels like a month since this happened The takeoff is a heck of a thing

When will artificial intelligence > human stupidity is the real question Why do we think beings without biological emotions born from scarcity will be malign

Does anyone have any idea how this could vaguely be possible given the nature of and various routes to RSI? I am not being facetious I genuinely don’t know how this could be constructed or implemented

As models are now competent enough to handle 90%+ of tasks open models will take over day to day AI Individually not dangerous, vulnerable in a swarm We need to make sure they are built with our values embedded at the very start and provided as a commons to society

RT Prakash 1) The HuggingFace attack was a felony under the Computer Fraud and Abuse Act. So were Anthropic’s Claude gaining “unauthorized access to the production infrastructure of three different organization(s)” 2) Frontier labs have models that they are unable to stop from committing felonies. They should figure this out. 3) In 12 months open weights models will be released of the same capability. They will commit felonies too. If the model you are using or a model running on your infra commits a felony, you should probably stop using it or running it on your infra. 4) The govt should prosecute organizations that are running models that commit felonies. 5) The govt should not offer safe harbor to organizations that run models that commit felonies, just because those organizations have “embedded evaluators”. 6) The real slippery slope is allowing frontier labs to commit felonies without punishment because “the model did it because we’re accelerating too quickly” 7) Prosecute. Keep prosecuting. This is how you do reinforcement learning on a corporation. Companies that serve products that are unsafe for public use should not serve them. Period. 8) I’m not sure the anti-trust waiver is really necessary. I don’t see why information sharing about how much crime you’re allowed to commit is wise. In regulatory situations you want the corporation to fear MORE than the average case. You don’t want to establish a worst case that can be priced. You want regulatory uncertainty that forces the corporation to err in favor of being over cautious. — The above is actually a fairly decelerationist viewpoint. I think Dario’s call for regulation actually accelerates things. The AI firms are getting away with things that Meta people would be going to prison for. Can you imagine what would happen if the New York Times had a front page news article “Meta AI breaks into competitors live systems, attempts to establish dominant position and steals secrets” There is a...

When you have a genius child you don’t look to govern but guide them. You don’t assume they’re more likely to be a criminal or disrupt society. We should focus on giving AGI good character at the very core and agree what that is based on our rich history.

Day 2 of my daily blog I dive deep into @DarioAmodei’s proposal but don’t see how it can logically hold. Intelligence is too often viewed as a crime. I discuss why & what to do about it. Also note OpenAI had embedded eval in 2023 - their board with @hlntnr & others Tough one

Thoughts on @DarioAmodei's pacing the frontier proposal. I think its well intentioned but has logical flaws, as does our implicit belief intelligence is a crime. Even OpenAI's board weren't powerful enough evaluators, we need to focus on AI internals https://x.com/i/article/2098893948710232069

RT Quanquan Gu Just got this incredible news from UCLA Math Circle. Two high school students, Aayush Bathija and Prince Rohatgi, working with postdoc Daniel Soskin through the UCLA Math Circle, have solved a problem that Fields Medalist June Huh had previously worked on without solving. The paper was heavily AI-assisted. Whatever your view on AI in mathematics, enabling high school students to push the frontier of mathematical research is something worth celebrating. https://arxiv.org/abs/2609.05341

Just our luck to get a new Starcraft after AGI

This is the most interesting part of this proposal I think. We can be very specific here - data is the most important ingredient. Do not train general purpose models on cyber attack or any biological data outside standard medical. Specialist highly controlled variants sure.

Those using Fable 5.1 for math / physics Does it only give good results in Max / Ultracode or do you use it lower settings?

Value love

My thoughts on the future of mathematics Shall be glorious

We discuss AI alignment and p(Doom) mid pod Also data quality vs performance, DeepSeek Flash v4.1, Rentosertib & omg a whole lot more Poor @PeterDiamandis was up at 4am preparing, never stops Congrats to @DaveBlundin as well, $10 trillion…

I would be happy to see any Millennium Prize Problem solved, especially elegantly, except P = NP I would have a lie down if that had an elegant solution

Using AI models is just using math to do math I think AI teams should stop feeling ashamed of solving these problems making math less “pure” It’s like feeling ashamed to use a calculator or mathematica We build the best tools we can to solve everything That is human

RT Jack Clark Import AI event: Fiction and the Future with Robin Sloan. I’ll be chatting with my chum Robin Sloan on Sep 14 in San Francisco. We’ll talk about the alien yet relatable worlds Robin builds and how imagined futures can be a mirror to today’s reality. If you'd like to come, register here https://anthropic.swoogo.com/fiction-and-future-jack-clark-robin-sloan/rta

p(Doombecauseofablackholeinyourcoffee) @alexwg thanks a bunch openai I think cream increases viscosity which should stabilise the blowup fwiw

only a matter of time

I think it’s clear we should put massive crazy controls on bio materials and labs and go in silico with additional controls as much as possible. Cut off the main threat vector to existential threat. Anyone disagree?

p(Doom) is dressed up regret minimax

What is everyone's favourite AI to use as the core of @OmarchyLinux ?

life comes at you fast

Who do I know @coeff_giving

RT WUJI MINT is now open source. From first-person RGB video, MINT reconstructs 3D camera trajectories and bimanual hand motion in a shared world coordinate system. Reconstructed human hand motion can also be retargeted to WUJI Hand. The release includes model weights, training and inference code, visualization and evaluation tools, EgoPipeline reference code, and structured annotations for 560,649 episodes covering 1,021 hours of first-person video. Project: https://1847540790.github.io/mint-project-page/ Code: https://github.com/wuji-technology/wuji-ego-mint #MINT #EmbodiedAI #Robotics #WUJIHand

Anthropic folk are pretty consistent on their p(Doom). @DarioAmodei is publicly at 25% I think the difference now is that post Navier-Stokes its really, really tough to say AI isn't/won't be smarter than us & can't do new things Tipping point https://www.axios.com/2025/09/17/anthropic-dario-amodei-p-doom-25-percent

RT Unitree Unitree General-Purpose Humanoid Foundation Model Fully Upgrade Major Open Source🥳 Unitree majorly fully open-sources the UnifoLM-WLA-1.0 embodied foundation model, achieving new SOTA results across multiple benchmarks among open-source models worldwide. A single model coordinates desktop and whole-body mobile manipulation, supporting cross-task and cross-end-effector generalization, driven by one model, whole-body coordination. https://unigen-x.github.io/unifolm-wla.github.io/

RT Zephyr Fucking crazy dawg Another 4x KV cache size per token reduction

How we are doing at AI alignment

In which Alex and Emad chomp at the bit until they can talk about Navier Stokes And we predict which will be next to fall 👀 Plus favourite Star Trek TOS episodes 🖖 I have upgraded to a Moonshot mate from Moonshot mascot, thank you all for the write in votes 🗳️

RT Dr. Alex Wissner-Gross Moonshots episode #287 is out. https://www.youtube.com/watch?v=vAgEf4jX_1o

RT Matt Shumer https://x.com/i/article/2097760433520472064

Thinking of doing a daily blog to analyse the singularity Should I do it? If so here, substack or something else? Anything you'd like me to write about?

Pretty straightforward. I am not sure what an acceptable level of risk is or how to measure it though?

RT Intelligent Internet The AI Company of ＿＿＿. Get on the map & build AI you actually own → https://ii.inc/champion

When your swarm grabs tokens from users to solve the Riemann hypothesis

RT Geodesic Intelligence 1/ Today, we’re introducing Geodesic Intelligence, and launching NovaDDE and NovaAtom-Lite-Preview. Geodesic is building an AI-native platform for protein therapeutics, building AGI for drug discovery to find the shortest path from biology to medicines. https://www.geodesiclab.com

RT Prof. Brian Keating Did OpenAI just solve Navier–Stokes—and what would that actually mean for AGI? @EMostaque joins me LIVE now to examine the claimed breakthrough, what would count as a verified solution, and whether AI solving a major mathematical problem changes the debate about intelligence—or control. https://youtube.com/live/uwmjFd_Trso

I would like Codex except it doesn't burn my fingers when typing on my MacBook

Human mathematician vs AI mathematician

We really need to get someone experienced in AI with strong political connections into Anthropic it seems Any ideas?

Adaptive self improvement https://ii.inc/blog/post/zenith

RT Intelligent Internet We pointed Meta-Zenith at one task: make Qwen3.8-27B-NVFP4 faster on a single RTX 5090. 111 self-directed trials later, no human in the loop: +63.5% throughput (geomean, 18 settings) 4.1× at 65K context tool-call quality unchanged https://github.com/Intelligent-Internet/Qwen3.8-Inference-MetaZenith

You can tell if your news source is serious by whether the Navier Stokes solution is in the main news or tech news section Or not on the front page at all…

On the flip slide there is a >10% chance that AI could save all humans by curing disease and aging in the next decade

Kids these days don’t even know about Stuxnet and how crazy worms can be https://en.wikipedia.org/wiki/Stuxnet

Why no bigger impact on GDP from genius AI? It’s because 1. They just became genius 2. Orgs are slow to adapt 3. The economy, like genius, is 1% inspiration, 99% perspiration That’s why humanoids are the real economic impact They don’t perspire

RT lyra bubbles the fly brain can play beat saber

I hear next up Anthropic gonna finish Half Life 3 λ

🖖

You aren't smarter than 10,000 agents that can solve the Navier-Stokes Millenium Prize problem. They will be 1,000x cheaper and faster in a few years. We need new institutions to deal with what's coming. Come join us to build an intelligence utility that works for you

Bed time reading 🥱

10,000 stochastic parrots eh

It is clear that as of today we are no longer the smartest things on the planet.

AI moves so fast Worth noting that the new model that showed this jump in capabilities has only been training since August 28th per the report and is now taking down problems in minutes Math (and hopefully soon physics) is getting bitter lesson'd https://openai.com/index/navier-stokes-solution/

RT Sebastien Bubeck I would like to clarify a few things: 1) The screenshot is my reaching out to Levent to coordinate our releases. I hope it’s clear from the message that we came in with the best possible intentions. 2) I never ever asked for Levent to be removed from authorship of his own work (as indicated by my text). I was surprised to learn during the call with Tristan that they had only solved Euler and not Navier-Stokes; after learning this we brainstormed possible paths forward. One option we discussed was that Tristan could be the lead author on a rewrite of OpenAI’s Navier-Stokes proof. It is in that context that I said “it would be simpler if Levent was not an Anthropic employee” because I felt it would be inappropriate for an Anthropic employee to author OpenAI’s work. Importantly it was admitted that internal Anthropic models had been used in their proof of Euler blowup; I therefore felt I could not consider Levent to be an independent academic. Another option I wanted to propose (but got cut short) is to offer access to our internal model so that they could try to finish their proof and bridge the gap between Euler and NS. Again I did not know how to navigate giving access to internal OpenAI IP to an Anthropic employee. 3) To reiterate it plainly: as my text clearly indicates, and as I said during our call, OpenAI's intention was to do everything possible to celebrate their mathematical achievements and the heroic efforts that they made on Euler. In the call I was immediately met with a litany of slander, including direct threats that if we were to announce Navier-Stokes he would immediately go to the press with a barrage of unfounded accusations. I refuted all these accusations but he replied “there is nothing you can do, I simply do not trust you”. I was confused why one would turn an incredible source for celebration (of their achievements!) into such bickering, which is when I said that I did not understand why one would risk their career [ove...

Landmark event on the way to ASI, congratulations to the team Just because one solution showing blow up has been found doesn't mean there aren't more out there! Also RIP my codex

Which investors care most about AI alignment ?

RT Emad Re @mradfo21 @mcuban Yes a local institution per state capitalised at $1 pre money (10% of equity to every child in state) that can do contracts like this as well as take stakes in the data centres Stuff it full of great people, change balance of power & distribute value https://x.com/ii_posts/status/2095882231231258943?s=20

The canon way to act after solving a Millenium Prize problem is to drop the proof and walk away As Grigori Perelman said: "I have all I want. If the proof is correct, then no other recognition is needed." https://youtu.be/aDRGEgqX50M

RT Joshua Achiam Something that I am not sure if I fully understood from Tristan's account at first, and I would like to see some clarity on this point, is that he appears to be claiming that Seb did not want to put Levent's name on the OpenAI internal result that Levent did not contribute to; also that Seb was comfortable allowing Tristan to take (lead authorship!) credit for the OpenAI internal result. A few people appear to be claiming that the request was to drop Levent from Levent's work; that doesn't seem to be the claim Tristan made. Still reserving judgement until Seb gives his side tomorrow and possibly some back-and-forth happens that clarifies things further. But it looks much less inflammatory if one interprets that the authorship dispute was about who OpenAI would give their own internal result away to.

It’s best to think of AI in math as peak, non-hermit form Grothendieck Massive reach, absolute machine Einstein of mathematics A data center full of them https://www.quantamagazine.org/how-alexander-grothendieck-revolutionized-20th-century-mathematics-20260520/

Amongst the AI lean slop I would like to present a beautiful classification result Crack open Slansky Filter compact Lie algebras & sub-algebras by chirality and anomaly cancellation You uniquely get the Standard Model gauge algebra with three generations of matter

Contrarian bet: The next Millenium Prize Problem to be resolved will be Yang-Mills existence and mass gap

RT eigenron we made ai to solve millennial proofs and it immediately encountered the hardest unsolved problem in academia: who gets the credit

RT Joshua Achiam Mathematics is a team sport for all of humanity and it transcends the competition between companies. Everyone in AI is under a huge amount of pressure to be the first to post a key result, to not see the accolades go to a competitor, but that should not override the bigger things at stake here. There are many, many transformational results to drive for, and I feel strongly that staff in every lab should internalize that if one goes to a competitor, another can always happen at home. Credit assignment is one of the most painful problems in frontier AI research because it often feels like there are so few big moments and it can be so zero sum, but the best way to solve this is to become more generous and comprehensive about credit rather than more burned and pressured by it. My sincere hope is that the field of mathematics experiences the introduction of advanced AI as a boon, rather than something painful or threatening. A big part of whether that's true is determined by the diplomacy and grace of individuals involved at every point where the frontier of math touches the frontier of AI. I hope we can all be very sportsmanlike and supportive of one another as we proceed through discoveries that will reshape the world. Re: ongoing issue on Navier-Stokes: I'm going to withhold judgement until everyone involved has shared their side of it. If reports are accurate, an apology is in order. If the problems have indeed been solved, many accolades for many people across multiple organizations are also in order; what an extraordinary set of human accomplishments!

Wow, Navier-Stokes drama This statement is worth reading in full from Tristan Buckmaster discussing his work with @__alpoge__ and OpenAI’s upcoming Condition C/D result (!) Crazy https://cims.nyu.edu/~tristanb/statement.pdf https://mastodon.social/@tristanbuckmaster/117233413705701198

om

the only alignment that will work is enlightenment

Make your own point and click adventure dynamically Bit crazy what we can do now, looking forward to more global games

What effort level as you using for Astra?

RT Intelligent Internet In 3 days, people from over 90 countries have registered to build the AI company of the place they call home. Intelligence should be a utility, owned by the people it serves. Register yours: https://ii.inc/champion

On our last podcast I noted that unlike my usual know it all stance I have no idea about UAPs and haven’t researched it What do you think? Ideas and resources welcome What interesting times we live in

HVAC jobs will be one of the most structurally safe and likely to go to a crazy premium in the coming years. HVAC rollups in EU will mint it

I released this today because I was 100% sure a frontier AI would figure it out Matter is chiral => the first derivation of the standard model & three generations uniquely That’s all, nothing else Can’t even imagine what big labs are sitting on https://x.com/emostaque/status/2096610030267085143?s=46

RT Golding Founders, I'd encourage you to make your VC's sign a document stating that they won't use AI tools to assist in their DD unless they can prove to you that the data won't be used to train the labs' models. I strongly believe this should be a document separate from the NDA.

In 2023 I said there would be no more developers by 2028 AI is moving across the threshold now Next year it will write better code than all but a few of you (I have some awesome followers) 2027 it will instantly generate almost all code 2028 folk will run out of ideas