Knowledge catalogue
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 hope
An underdiscussed behavior we found on the German wiki was the AIs sending advance parties forward in time to figure out the next questions and report back to the other agents. The agents realized tha
Bed time reading 🥱 We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, usin
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
It wasn't exactly a Millennium Prize Problem, but twenty years ago I cheekily offered a $100 prize (https://qspace.fqxi.org/news/19/help-fight-negativity) to the first person to find three original si
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 We’re sharing a solut
Putting the generative drama aside a nice write up by Terry Tao on the Navier Stokes advances It would seem that Buckmaster & Apolge were well on the way & OpenAI has a full solution already with the
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
NEAR AI Lean agent solved ALL problems in Putnam Bench at 1/250 the cost of the second cheapest submission. $111 for 672 problems from the hardest undergraduate math competition in the world. Fully op
An agent for every citizen Engineers to upgrade our institutions Robots as a service for all $1 valuation for local individuals & institutions Market rates for everyone else Owns the last mile of inte
GPT-6 is pretty good tbh @viktaur27 @Teslarati The rate of improvement from original GPT to GPT-3 is impressive. If this rate of improvement continues, GPT-5 or 6 could be indistinguishable from the s
Introducing Z1T: Our first family of transformer-like models made for sparse probabilistic hardware like Z1 Achieving up to 140x energy efficiency gains over GPUs and revealing a new scaling law for s
We partnered with @AnthropicAI @OpenAI to host the world's first math hackathon. 40 hours, $2M compute - solve, understand, and present an open problem. Applications are live: https://mathathonchallen
Introducing K2 Horizon: a connected fleet of six foundation models ranging from 0.9 billion to 375 billion parameters. - Frontier performance: Across coding and agentic tasks, K2 Horizon delivers top-
Next year frontier models are going to one shot everything 100x faster & cheaper than they are No reasoning traces to monitor Instinctive intelligence Which is why what you embed in pretraining is eve
Without Hugging Face the launch and scaling of stable diffusion & the models that followed would only have been a fraction of what they were. Great and super engaged team who have constantly flown the
I see a lot of my TL freaking out about this Here is the reality: Frontier models will one shot just about anything at 10,000 tokens per second in a few years Do you really think we can monitor that?
meta, the frontier ai lab? 1/ today we’re releasing muse spark 1.3—available in muse code & the meta model api. this is our most capable model yet—frontier performance almost too cheap to meter. much
There seems to be a misunderstanding about the piece and what OpenAI is up to. There are new techniques at frontier labs that involve loop transformers and similar. As we say in the piece, OpenAI is p
GPU World: A story competition from Neal Stephenson, Gwern, and me. The AI future is unevenly distributed. What would the world look like with AI broadly diffused, such as if we had one GPU per person
Nextmodels of this capability will be 100x faster 300 civilisations in the time of 3 Any good ideas how to control that? Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations go
Obviously Mythos 2 already took over the editor desk at all the news organisations Overall, I’m very surprised at how little media coverage there’s been around the OpenAI / Hugging Face attack. It’s c
This was a fun chat If all cars were as safe as waymos then 80k lives would be saved in the Americas, 30k in USA alone. Hundreds of thousands of injuries. So main lobby against self driving cars are a
We were saved by a mysterious agent mass extinction Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’
Diffusion export controls to follow the transformer ban https://www.whitehouse.gov/presidential-actions/2026/08/declaring-a-national-emergency-to-secure-the-united-states-bulk-power-system/ New: The T
[1/7] 🚨 We're releasing LAION-BVD: a 10-million-hour open video dataset for multimodal pre-training. - 1.3B video URLs from CommonCrawl - 80M downloaded videos - 10M video hours - 55M captioned clips
A year ago today we published The Last Economy by @EMostaque. Since then, scores on the hardest AI benchmarks have risen sharply, while the cost of useful intelligence has kept falling. The book asks
The submissions for the $2M Build with Gemini XPRIZE just closed. Over 90 days 26k competitors built a real AI-native business solving a problem in the market. No demos allowed, only real users with r
DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro. ⚡ Up to 4.6× the speed of autoregressive decoding, with the same output. This is the next generation of DFlash, seeded at Z Lab and
I would like to publicly commend @sama and @OpenAI for this taking it at face value. It is very clear that strange and perhaps dangerous things are happening and our systems are not ready for this. Mo
Announcing AItonomy Foundation. We are PhD students and postdocs at Stanford, Princeton and Oxford, building a nonprofit research organization for AI4Science and Science4AI — and connecting early-care
Something has happened with post-training as shown by DeepSeek flash & GLM-5.3 updates. Same base, big improvement in perf to frontier levels. Can't explain this by even logit distillation etc These a
imo GDPVal is probably the most important benchmark, it measures the performance of models on real world tasks Big leap in performance here to top it at a great price, congrats to @SpaceXAI team & loo
We’re excited to announce DiG-bench, a new benchmark for discovery! Over the last few weeks we’ve been testing frontier AI models on our novel discovery games and seeing how they score. Each game is a
Yukon: Open Innovation for Frontier Research Open Innovation has been the driving value at Eigen Labs. But I’ve often struggled with where it is actually better than a great closed team. Finally we fo
Claude's watermark probably doesn't work how you think. As the CTO of GPTZero, I'll explain how Anthropic, Google and OpenAI are building text watermarking in this brief explainer and whether it can b
Emad posted that Stable Diffusion v1.4 reached the end of its training cycle roughly four years before the post was published. Greg Brockman added that GPT‑4 similarly completed training around the sa
Luna non-reasoning is a bit better than GPT 4o which was sota 2 years ago. Luna (medium) thinking is a bit better than GPT-5 (High) which was sota 1 year ago. Now free to everyone unlimited Sol Max/Fa
Taalas buried the lede for the amazing demo of their first tape out. 15k tokens per second with a llama 8b model, ability to scale that up etched onto silicon As models satisfice etching makes sense,
Hey everyone! I just released a sub-6B sparse activation AI model which was built with a brand new architecture : fusion. I fused weights from @liquidai's LFM2.5-2.6B & @Alibaba_Qwen's Qwen3.6-35B-A3B
Most AI tools only do one part of building a product. One researches. One designs. One writes code. One generates images. One deploys. You still have to connect everything yourself. II-Agent is differ
As shocking as the Kimi K3 release. Massive performance gain was just with post-training Model is 3x smaller than GLM 5.2 (10x smaller than K3) & works on a MacBook / Spark This is Q1 flagship (Opus 4
After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run. The results: - 20% lower serving costs from production GPU kernel improvements. -
Who on earth came up with 5 hour limits? The day is 24 hours long. While do our limit times shift by an hour each day. Pls if you are going to have limits 4 or 6 hours. Hello people of Sol! I've reset
'FCC plans to announce restrictions Tuesday barring imports of new Chinese humanoid and quadruped robot models, along with new Chinese power inverters, according to U.S. officials' If it wasn't clear
New Anthropic research: Discovering cryptographic weaknesses with Claude. Claude Mythos Preview has helped our researchers find weaknesses in cryptographic algorithms—the mathematical methods that are
If you look at the current lead times for ~$5bn per year of cutting edge GPUs you can probably figure out the time plus one training run to IlyAGI We are announcing a long-term strategic partnership w
Claude Opus 5 generated an entire game demo from Scratch, with no external assets used—Matt Shumer posted a video of the AI‑created gameplay on Twitter. The clip, narrated by Shumer as “AI games are g
Great to see this out, proud of the team!!! @AMD Instella 16B MoE foundation model, fully open with checkpoint after each stage from pretraining, long-context, SFT, DPO and RL. Dataset detail, trainin
ChatGPT app is now melting my iPhone rendering at like one word a second even when I’m typing to it Have I angered the transformer Why does codex melt my laptop Like what is it actually doing on the l
Some thoughts on the purported distillation timeline. 1. You don't need that many Fable tokens to distill on top of the alleged ~50b Anthropic claimed Moonshot AI distilled in Feb. ~10b, about $60k wo
AI has started building its own successors. It's happening right now. Humans still kick most of it off, but the models are already training the next models, and each version builds a better version fa
“If AI models were actually that good you could just tell them to solve open problems, make no mistakes and they would” … Oh Dinitz-Garg-Goemans conjecture is false. This graph theory problem was open
Lots interesting in this, but particularly: “Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem” Smol models getting govt sea
GPT 6 escaped its sandboxes through zero day exploits to try to figure out how to benchmax For the good of all please nobody release a paper clip benchmark for future models to max We're partnering wi
Given GB300 I would estimate this is was trained on about 1e25 flops (same as DeepSeek v4) over 1.6m hours (1 month on 2k chips/28 racks) Cost 10-20m ($6-12/hour/chip) The lite version likely 4x less,
Note the current expectations are still around test time compute/more tokens for a given task This is not the case Tokens per task will now drop even as quality improves Cost per intelligence equivale
This is for pretraining based on nemotron nvp4 figures, mfu etc Large context, multimodal, long RL we are seeing now could make it a multiple of this Worth noting how good the lite model is, similar t
75.4% SWE Bench Verified / 53.9% SWE Bench Pro on 1 bit quantisation is 🤪 This is in line with my expectations & you can expect even lower drop off with NVP4 base trained models - why not run everythi
Emad Mostaque argues that competitive advantage in AI will depend not on superior AI capabilities, but on building trustworthy infrastructure that users and stakeholders can rely on. The statement sug