AI, ML, Brain and Space enthousiast
Lisbon, Portugal
Joined June 2010
- Tweets1K
- Following471
- Followers235
- Likes2.3K
Thierry retweeted
Ten years ago, AlphaGo’s Move 37 shocked the world.
It wasn’t intuition alone that produced it. AlphaGo could search possible futures, test its instincts and reason about what would happen next.
In a new piece for @techreview, I argue that today’s most advanced AI systems are still missing something fundamental.
LLMs are remarkably capable, but generating longer chains of thought is not the same as genuine reasoning. They typically have no explicit, inspectable record of what they know, what remains uncertain, what evidence supports a conclusion or whether genuine progress has been made.
This is why I recently left @GoogleDeepMind. I believe we need a fresh approach to machine reasoning, drawing on some of the architectural lessons from AlphaGo.
If AI is going to produce trustworthy and genuinely novel insights in science, medicine and beyond, we need systems whose conclusions arise from an auditable process of evidence, inference and belief revision.
Thierry retweeted
“One of the reasons why I am in favor of less government is because when you have more government, industrialists take it over, and the two together form a coalition against the ordinary worker and the ordinary consumer.”
— Milton Friedman
Thierry retweeted
A Severe Misalignment of AI in Mathematics
New blog post by Terry Tao and a new anti-AI declaration signed by 25 Fields medallists.
I'm not sure what the intendent effect should be. The ban of AI use in mathematics and science in general because of "math community"?
Fortunately math community is broader than just math academia, and thanks to AI, booming like never before.
Also this "mass production of true/false statements" is such a wrong description. As if there's no proofs by AI that you can study, deepen your knowledge, build upon, engage. The "AI slop" is still better written than what 90% of mathematicians write.
Mathematics doesn't belong to math academia. It is a tool to understand the world and help other domains with necessary tooling. AI boosts the understanding and engages wider community.
Swan song of academia.
Link: terrytao.wordpress.com/2026/…
Thierry retweeted
We can finally talk about it:
We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.
We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.
Thierry retweeted
If you're ever giving a talk on how difficult it is to do data anonymization well, this is your lead slide.
Thierry retweeted
Our general approach is to automate the experimental loop. We think this approach is broadly applicable across many different fields of science and engineering. We’ll initially focus on ML research and engineering, but believe the approach can help with important subproblems in nearly every one of the fourteen <at>NAE Grand Challenge problems. We think doing this well requires strong expertise in machine learning as well as large-scale systems.
Thierry retweeted
I finally have all 1.2 million raw image files from my latest mission to ISS! Here is a sample of one of my favorite Milky Way photos, taken from the Cupola with Nikon Z9, Arri Zeiss 15mm lens, T1.8 with custom sidereal drive that cancelled out star motion relative to our orbit.
Thierry retweeted
Today, we are releasing Inkling-Small.
Inkling-Small achieves comparable performance to Inkling at a quarter of its size. It features 276B total parameters, 12B active. We are making the full weights available.
thinkingmachines.ai/news/ink…
Fine-tune it on Tinker today, or chat with it in text, image, and audio on Tinker Playground.
Thierry retweeted
Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set.
[schema] makes an LLM think like a physicist. 🧵
Thierry retweeted
Yesterday, we made GPT-5.6 Sol Ultra generally available. Today, we're sharing that it produced a proof of the 50-year-old Cycle Double Cover Conjecture using 64 subagents in just under one hour. We're sharing the prompt and proof below. We're excited to see what you all do with Ultra!
Thierry retweeted
Bro I'm so sick of pretending this isn't weird.
The internet spent 20 years creating tutorials, open-source projects, blog posts & answers for free.
AI companies turned all of it into products worth billions.
And now the same people who created that knowledge are being told they're replaceable.
We built the library.
Someone else started charging admission.
Thierry retweeted
LLMs learn by predicting tokens. World models (JEPA, data2vec) learn by predicting their own abstractions. Which needs more data? For data with hidden hierarchy, we prove the gap is exponential. arxiv.org/pdf/2605.27734
Thierry retweeted
Introducing DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretation
pub.sakana.ai/diffusionblock…
What if we didn’t have to hold an entire neural network in memory to train it?
Standard neural net training optimizes all parameters jointly. As a result, the memory required during training grows linearly with the depth of the network.
In our #ICLR2026 paper, we propose DiffusionBlocks, a principled framework to train networks one block at a time, drastically reducing memory requirements while matching end-to-end performance.
With DiffusionBlocks, we split the network into blocks and train them one at a time, so you only need memory for a single block.
How? We explicitly assign each block a role: to move the representation a little closer to the target than the block before it did. That role turns out to be precisely what a diffusion model does, step by step. Each block only needs to optimize its own objective and can be trained independently.
We validated this across five different architectures:
• ViT
• DiT
• Masked diffusion
• Autoregressive transformers
• Recurrent-depth transformers
In each case, performance is competitive with end-to-end training while using a fraction of the memory.
This perspective also extends naturally to recurrent-depth (Looped) transformers, which apply the same network iteratively and normally require expensive backpropagation through time (BPTT). Viewed through DiffusionBlocks, we can replace those multiple iterations with a single forward pass during training.
Read our paper and code, to learn more.
Paper: arxiv.org/abs/2506.14202
GitHub: github.com/SakanaAI/Diffusio…
🐟
Thierry retweeted
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
Thierry retweeted
"Coding is [0%] solved".
Thierry retweeted
there’s a dude on youtube making a video with a burst blood vessel in his eye. the title is “software developer driven to insanity by 2026 job market” and he’s pointing at his eye as evidence of the stress he’s going through.
garrett’s a rather normal dude and there were no obvious tells as to why this guy has not been able to find a job for 9 months since he was laid off in july.
but there was a hint.
at one point he describes how he went really far in an interview process, and he was absolutely sure he was gonna get the job. they hire someone else and the dude is incredulous. he asks them wtf!? they show him the winning candidate entry.
garrett used 50% ai and 50% manual coding. he was sitting there being thoughtful and deliberate and properly naming functions and variables and designing modular architectures. meanwhile the winning application was a dude who used 100% ai and his app had a shit ton more features. and he delivered it faster.
so, that’s the kind of dude being hired now. garrett didn’t get the job. but someone else did. so the job is there. the job was open. just not for you.
the thing getting you hired in 2019, like obsessing over code quality and maintainability and best practices, will get you fired in 2026.
the market has spoken. if you want to land a dev job, make more slop.
they want more features done faster at a lower cost.
your job is not to argue. it is to provide the thing they are seeking.
pick 3 companies you really wanna work for, then build their entire app in a weekend as your cover letter. applying to doordash? build their 3-sided market place in 2 hours. applying to google? build your own search engine that does a 2-hour web crawl.
reading code is a dead thing. no one reads their own code anymore, let alone yours. they will run your app. and they’ll judge you off that.
build more slop. build a slop empire. imagine more slop. be the slop. slop slop slop.
get the job. make money. give your boss the slop he wants. don’t argue. slop is peace. slop is strength. inhale the slop. exhale it.
oh you wanna be an artist? cute. go code on a city street like a bucket drummer. you code by hand for tips now. otherwise, off to the slop factory.
enjoy it! enjoy the sloppagaden! make a little money!
