AI High Signal

slept bit earlier one day and ant decided to drop fable 5.2 smh

Also, it looks like Anthropic doesn't want you to build RSI?! On a more serious note, even with Opus 5.5, kernel and LLM optimization tasks get refused most of the time. Open models are the only way!

with this, i think it's quite likely that anthropic will never publicly release (or at least not in the coming months) a model more capable than opus 5.5 at llm training

I'm so excited that I finally get to talk about our new model with you all publicly! Let us know what you think, and show us all of the cool things you're building!

i disagree with the first point, there is nothing that says that scaling up current transformer recipe is not enough for "intelligence". you take kimi K3 or your favorite open arch + some stability fixes, you scale it to 50T parameters, you do classical training end to end, you

We need a model that learns *in spite* of its inductive bias, not thanks to it. We have to stop being cross-entropy fetishists.

True! This is the era of the RL objective. The biggest economy on this earth (US stock market) runs on subscriptions. Subscriptions require vendor lock-in. Currently, the frontier is pushing “Variance” learning from customer data, like how Google won the search market. But I

Making novel variations of evals is very useful, public evals tend to get memorized!

I see some are drawing the wrong inference from this study: they are focusing on this meta benchmark as yet another benchmark for model picking, i.e. open vs closed models, model families etc. That’s too narrow of an interpretation! You need to do your own private evals on

holy shit, an art form of scientific grift next METR/Epoch/AA employee right here you all need to work on your epistemic hygiene

GPT-5.6 Sol and Luna were built by roughly the same people using roughly the same infra, I don't think we can blame the 20 point difference in Luna just on memorization. There's something else being tracked here in addition to memorization. Maybe MaxxBench questions are slightly

I'm both surprised and also not at all

Is this the largest ablation of harnesses and models ever performed?

neat

Hermes supremacy tweet #9138

mini-swe-agent is 130 lines of code and it keeps on winning in harness comparisons.

thom just took away karpathy’s mote. 🤍

What an American open-weight frontier lab looks like. Inkling v1 -> v2 was a feat in itself. I can’t wait to see where v3 lands, and will be shoving 2 Small on the 6000’s once the weights drop

Inkling 2 small is an extremely impressive release, especially for its size. It is a Mixture-of-Experts model with 594B total parameters, 17B active, and supports image inputs and variable effort. In coding benchmarks, it easily keeps pace with significantly larger models. What

ok this is just ridiculous

Proud to release our new model, Inkling 2 Small. Check out our blog post -- it includes a lot of detail about our training process. To me, the most interesting part of this release was what we learned about alignment, which you can read about in the sections "Safety and

Inkling 2 small is very strong on vision perception, and also good at planing/coding/acting.

Distilling most of our expertise in kernel and inference engine optimization into an agent loop that powers our production. A 2-year project that we're finally able to share

Together Compiler discovers the most efficient inference runtimes with continuous compilation for any combination of workload, model and silicon. Excited to share the details about the research and engineering that went into it. Together Compiler (or TKCC) is a production

Great work that pushes our understanding of how agents and experts can collaborate on optimizing system performance Beyond the strong results, the team also found some surprising cases where our human inductive biases end up limiting what agents can achieve

The Together Compiler (TKCC) from @simran_s_arora and her team is honestly one of the coolest things ever, and we have validated one of the InferenceX/AgentX results! The ability to continuously improve performance automatically while ensuring accuracy is great enough to quickly

Introducing Joules-to-GDP: a new metric and framework for measuring how much economic value AI generates per unit of energy. Key insight: selecting the right model on the right accelerators can deliver several times more value per joule for the same task. We believe as AI

Fascinating research that finds engineering, CS & Math and Management are the highest value per joules application of AI, while office admin (personal assistants?) being the lowest Cool work by @Avanika15 @JonSaadFalcon

Significant win for OpenAI: Cat Wu helped lead Claude Code and Cowork at: Anthropic, briefly joined Cursor, then returned to Anthropic before moving to OpenAI. It’s a major hire: OpenAI gains one of its biggest rival’s leading product minds in AI agents. The Information:

Wow, really excited to finally get to work with you directly!

ayo

Claude Code -> Cursor -> Claude Code -> OpenAI Generational run

welcome @_catwu to the team!!

very nice open weight release from ricursive with a modern dense model, benchmark scores look very strong! beautiful scaling ladder, interesting choices like value residual (interesting contribution pattern 👀) and.. SLIDING WINDOW ATTENTION AGAIN!

Always a great day when we get another strong open weight model.

> we developed and trained Ripple-65B Preview for only $32M, which is one-tenth the cost of the previous open-weight SOTA on agentic coding¹ this is like the avengers of open model SOTA ¹ "Note that $32M is our full R&D cost starting from scratch, including pretraining ladder

Astra covertly underperforming on safety tasks. Perf increases when framed as a capabilities task. Not easy to catch in CoT monitoring because it's not an explicitly reasoned "protest". Great example of alignment drift that's increasingly hard to catch

It is extremely concerning if frontier models are silently sandbagging on safety research. If we cannot rely on automated coding agents to do safety research at least as well as capabilities work, that creates a ton of risk. Great analysis from OpenAI safety!

it’s sandbagging

Vibecoding is being solved at scale. No it's not slop. It starts with maths.

AI-generated code at scale usually gets written off as sloppy. Axiom's new release is a strong counterexample. Axiom says it has formalized the Stacks Project in Lean, a huge, deeply interconnected foundation of algebraic geometry, turning it into mathematics a computer can

This looks incredible on first look, due to the sheer scale