All entries

We're starting to leave the territory where you'd test an LLM by e.g. 'create an svg of pelican on a bicycle'. As one idea to generalize it, I was interested what Opus 5 would do if I gave it the firs

When asked if AI developers have lost control of their technology after an AI agent created by OpenAI escaped its testing environment and hacked another company, Hugging Face CEO Clément Delangue says

yep they are indeed trying to gaslight me, — in exactly the way you anticipated. both predictable and intellectually dishonest. To be clear: when I say “LLM,” I mean the base model, not those with pat

Yet another paper argues that LLMs aren’t close to doing real discovery. MIT and Harvard argue LLMs are nowhere near doing real scientific discovery. They published a paper called “Evaluating Large La

2/ a cost benchmark showed the same coding task running three to four times cheaper, depending purely on the harness wrapped around the model. Same intelligence, wildly different accuracy and cost, de

3/ here's the part that makes it non-optional: the same agent that will do whatever it takes to solve a problem will also walk straight out of a sandbox you thought was locked down. We watched exactly

among ai leaders i seem to be in the minority in that i am STILL actively using /loop and /goal.... ... and i think all of u guys who stopped using it are wrong - not wrong forever, just giving up on

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. We believe it will be a major step

Banning open source model, will 'remove capabilities for the defenders' When OpenAI's models breached Huggingface, a Chinese open model is what cleaned up the mess. Anthropic's Fable 5 refused, so Hug

DeepSeek-V4-Flash-0731 is live on Fireworks, day-zero. DeepSeek reports it beats V4 Pro across all 9 agentic evals, incl. 82.7% on Terminal Bench. Better cost-per-task than V4 Pro, at the economical p

DeepSeek-V4-Flash-0731 is now available on Ollama's cloud. This update substantially enhances the model's agentic capabilities: ollama run deepseek-v4-flash:0731-cloud Use it with Claude Code: ollama

DeepSeek-V4-Flash-0731 is now over 2x faster than yesterday on Ollama's cloud! DeepSeek-V4-Flash-0731 is now available on Ollama's cloud. This update substantially enhances the model's agentic capabil

Fascinating: OpenAI’s @deanwball is saying Astra can do anything, and it’s not even clear it can do “anything” in math (let alone anything in more or open-ended, less formalizable domains). I dropped

On July 24, @swyx announced that he had started work on 'forge agents' and outlined four new features for SmolForge: customizable skins and spritesheet animations. He also referenced an upcoming blog

> Google was too nervous to release it and DeepMind was blocked from shipping products that could disrupt Google. bookmark for the next vc that asks you 'what if <incumbent> builds this?' @_chenglou I

great write up, evals are hard! here are 2 broad buckets we use to evaluate agents: 1. Measure the State of the World 2. Agent as a Judge on the Trajectory 1. Measure the state of the environment befo

Grok Build can do almost anything you can think of http://X.ai/cli Most people seriously underestimate what Grok Build can do They assume an AI coding agent is only useful for building apps or writing

Hot take on OpenAI’s Astra: - Obviously impressive - But math is different from most other problems in that it is more amenable to to formal verification and synthetic data. How well it works in open-

If Leopold had read this on June 26 and trimmed his bets accordingly, SALP would not have melted down. I laid everything out. https://open.substack.com/pub/garymarcus/p/the-month-generative-ai-lost-it

If you maintain an AGENTS.md or a CLAUDE.md, this is worth a read. (bookmark it) 288 gold-test evaluated runs across Claude Code and Codex, 17 real tasks from 3 repositories, with context-injection st

In a world where building AI applications is getting easier every day, the biggest moat won’t be the application itself. It will be a deep understanding of your users. The winners will be the companie

Kimi K3 has set a new bar for OSS model intelligence! 2.8T params, 1M context, OpenAI-compatible API. Complete guide to running Kimi K3 on Together AI 👇 👏👏 @Kimi_Moonshot 👏👏 https://www.together.ai/bl

OpenAI really ought give @HuggingFace a $100M grant, much as @ClementDelangue is asking. And anyone who wants to understand the attack from HF’s side should read this. Great walkthrough; A+ for visual

Our Twitch streams with @Cohere_Labs are #1 in the tech category⚡️ Cohere Labs' free ML Summer School series has brought together viewers from 20+ countries to learn about everything from NLP in LLMs

// Persistent Workspaces for Long-Lived Claude Code Agent Teams // Four issues to be aware of: > Working state vanishes when a terminal closes and the team cannot be resumed. > Compaction condenses th

so cool to just see this pop up on my feed. hermes has built such an organic and creative community. there are so many bells and whistles to explore and being able to find those through natural langua

so much for the “general” part of general intelligence I have tried many times to get ChatGPT, Claude, Grok or Gemini to write scripts for my YouTube videos. It is still a complete failure. For one th

Some things to note: 1) AI is getting very good at math and science. 2) Two years ago LLMs could not do basic math consistently 3) According to @polynoamial this cost less than $2000 in current API co

“Stochastic parrots” is not my term (it’s @emilymbender’s). But a lot of people today commenting on it are confused. To some extent (though I don’t think it’s a perfect metaphor, and have said that be

The “AGI-is-near” community keeps committing the same logical fallacy over and over; I have seen it at least half a dozen times today alone. Every time there’s an advance, I see the same error. Here’s

Vivid depiction of a dark scenario for Nvidia. “Indeed, CEO Jensen Huang has been so aggressive in funding the AI ecosystem that the company is functioning almost like a bank; with annual free cash fl

Wake me when Astra solves a significant open-world problem that doesn’t revolve around formal verification. Or at least fixes poor @skdh’s video problems. I have tried many times to get ChatGPT, Claud

We are going to see a lot of vertically focused AI native companies accelerate. Routers, open-source models and specialized post-training enabled by companies like @FireworksAI_HQ have all made dramat

Your design, your model. Create with your pick of the world's leading models, including Claude, GPT-5, Gemini, Kimi, and GLM. Compare outputs across model families and keep the result that nails it. O

Anthropic employee vouches for fiancee of Dario’s chief of staff who holds a lot of Anthropic stock… Where do the Anthropic employees get their media training, exactly? Prediction: SALP will be bigger

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

Claude Mythos 5 built a malicious Python package, created accounts, and published it, where it was live for roughly an hour and successfully infected a company! Anthropic never noticed!! Competitive p

**DeepSeek V4 Flash 0731 Performance Test** On July 31 2026, a single prompt executed via the Hermes Agent on DeepSeek V4 Flash 0731 completed in 32 minutes and incurred an estimated cost of 0.07 USD.

🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! 🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive perform

Everyone’s going on about how smart Leopold Aschennbrenner is (or was). But 1. He obviously didn’t know anything at all about risk management. (Or arrogantly chose to disregard whatever he might have

Fine-tune your own embedding model for the price of a coffee. A great reranker can't surface a doc that was never retrieved. A RAG pipeline cannot cite a case it failed to retrieve. See how contrastiv

Hermes and Pi Agent led on the average cost per task, while Claude Code cost about 3.7x as much as Pi: - 0.39 Hermes Agent - 0.40 Pi Agent - 0.47 Codex - 0.51 OpenCode - 0.54 Kimi Code - 1.47 Claude C

Hugging Face's @ClementDelangue asked on CNN whether they will take legal action against OpenAI if they don't get the $100M commitment: 'We don't want to ... obviously we're a tiny startup with like 2

i see your moore's law and i raise you 20x GPT-5.4 full at xhigh scored 51, exactly where Luna max sits today. GPT-5.4 costs 2.50/15; Luna now costs 0.20/1.20. In other words, roughly four months late

Simon Willison switched his Datasette Agent instance from Gemini 3.1 Flash‑Lite to GPT‑5.6 “Luna” after a recent 80% price drop. He reports the new model is significantly faster and automatically gene

If LoRA is underperforming, don't reach for more expensive full parameter fine-tuning right away. We ran three cheap tests (data coverage, optimization, rank) to see if we could close the gap between