All entries

I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached. TLA+ also works well. I sometimes combine L

I'm happy to announce that I've joined Hugging Face. What started as a personal project back in February is now something I get to work on full time. Local AI has grown explosively this year. I've sai

Impressive paper showing how much the harness changes a coding agent's results. Harnesses do play a huge role in what you are getting out of the models. GPT-5.5 was run inside both Claude Code and Cod

In our evaluations, we see that GPT-6 Sol reads about 17% less context than GPT-5.6 Sol to reach the same score, and rarely leaves a repo with failing tests after its last edit. GPT-6 Luna’s biggest g

In the past, markets would flush out bullsh!t companies by requiring them to, eventually, be profitable. What’s annoying about OpenAI and Anthropic is their evasion of that mechanism. Too big to fail,

Interesting work on long-horizon research agents. PrimeScientists can decide where a research agent spends its budget. Achieves 10.3% more reward with 50.6% fewer research attempts, under the same bud

Kimi K3 is now on Amazon Bedrock! Run coding, document analysis, and extended agent workflows with Bedrock's access, encryption, and auditing controls. Explicit prompt caching supported. Start buildin

Last October, the Center for AI Safety (CAIS) reported that AIs could automate only 2.5 % of randomly selected remote projects. Updated results from the Remote Labor Index, posted on September 22, 202

Mars Sample Return mission is dead in the latest funding agreement. Which means China will likely beat the US to returning the first samples from Mars. Perseverance has collected the best samples it c

MiMo-V2.6 is 'simply' the best (for now). Despite its simple architecture design it's currently No.1 in the open-weight benchmarks (weighted average). With 'simple,' I mean a classic Grouped Query Att

Must-read papers of the week ▪️ JEPA-Anything ▪️ Modality-Autoregressive World-Action Models ▪️ In-Context Robot Learning with VLM Agents ▪️ Dream-RSI: Recursive Self-Improvement through Evolving Worl

🚨NEW: @JTLonsdale & @SquawkCNBC clash on AI and wealth inequality. Joe Lonsdale: 2030s is the disinflation decade... we're all going to be really wealthy if we don't break this. Joe Kernen: Everybody'

New on Dedicated Model Inference: canary rollouts. Upgrade the model behind a live endpoint without downtime. Traffic moves from your current deployment to the new checkpoint in gated steps (default 5

now on replicate: “Hot dog or not?” at 0.4 sec 🌭 jev-style decision model for image [glance-qwen3-vl-4b] now on huggingface and replicate: Demo: https://glancevlm.replit.app Replicate: https://replica

Opus 5.5 is the best frontier model for parsing tables in PDFs. We ran it through ParseBench, and it scored 93.9%, a 7%+ increase over Opus 5, and beating Fable/Gemini/Astra. It's comparable to LlamaP

On 22 September 2026, Allie K. Miller posted on Twitter that the Opus 5.5 model led her to curse at Claude less frequently. She notes that while reducing profanity toward another AI is an undesirable

Over the last few hours, some requests to the deepseek-v4.1-flash model on Ollama were charged at an incorrect rate, leading to higher usage consumption than expected for certain users. Sorry about th

People got upset at me for saying this, but there are no companies outside the US and China even *trying* to build frontier AI (perhaps Korea and UK are close), despite marketing claims. Governments s

People outside the AI labs should have a real say in how this technology develops, and a clear way to judge if it's happening safely. Standards should help prevent the concentration of power, includin

Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe. GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to sup

Victor M (@victormustar) tweeted on September 22, 2026 that the MiMo V2.6 release now includes an official Distill‑Qwen‑9B variant. The tweet expresses enthusiasm for this new model iteration and remi

Recommended if you work on agent memory. Retrieval is the hardest memory problem in most harnesses because agents keep pulling stale context. AML tests this with a coding track of 150 software tasks,

Run GGUF models directly with transformers. This work brings ggml's Metal kernels to the transformers ecosystem, increasing compatibility and performance. More info below Millions of GGUF downloads la

The fastest PDF-to-Markdown parser just got faster. ⚡️ With LiteParse v2.14.6, text-based PDFs parse about 25% faster. Across realistic documents, LiteParse processed pages at 2.8ms/page and 1.5× fast

First model of Laya was produced with a single overnight training and subsequent fine‑tuning run, taking only 15 hours from the start of training to publishing the model. It was released on Hugging Fa

The longer the AI psychodrama goes on, the more it’s clear we are dealing with a conventional business bubble. What’s different this time is the lengths to which the companies will go to put off the r

This is why people who call themselves decent vote for indecent policies decade after decade after decade. Because it never hurts them. Theirs is a monstrous indifference, because empathy is for the l

We're excited to release the State of Generative Media Report, Volume 2. Generative media has moved from experimentation into production and that shift looks different across every industry. Built fro

AI can be a really wonderful tool for exploring topics far from coding. I had Claude Fable 5.1 put together an annotated guide to Eliot's poem 'The Wasteland,' with multiple pathways through the poem,

and here’s the grok 4.7 model card. a few jumps vs 4.6 that stood out: - Terminal-Bench: 20.3% → 38.0% - SWE-Marathon: 31.9% → 46.0% - HealthBench Pro: 48.5% → 56.7% - Legal Agent: 15.8% → 19.6% - EEB

Banger paper on self-evolving ontologies for agents. You just can't go wrong with implementing an ontology layer for your agents. This paper shows exactly why. The show that GPT-5.5 gains 26.7 points

BREAKING: Grok 4.7 dominates the Legal Agent Benchmark. It scored 19.6% on realistic, long-term legal tasks, beating every other model tested. That is nearly 3× Fable 5.1. Another major win for Grok 4

🚨 BREAKING: US Treasury Secretary Scott Bessent declares OpenAI management personally responsible for HuggingFace hack > 'The Hugging Face incident, that is the responsibility of the OpenAI management

coding agents are moving fast from prototype to production. the infrastructure question is what's left. join @parthsareen from @ollama and @zainhas Hasan from Together AI at @AIconference for a breako

Document classification and splitting idea using @typesafeai Jev based on LiteParse, and LlamaParse. Classify: categorize probabilities, and review flags. Measure: separate OCR and decision timing, pr

Finally, a company brain that actually works! Codos starts with the people doing the work. It interviews employees to find which tasks can be automated, then deploys the automations and keeps learning

On September 21 2026, Yohei reported results from a hard “insect‑identification” benchmark where model accuracy spanned 23 points across the evaluated systems. The Qwen 4B + Glance model scored 0.96,

Found a great production use case for Jev. I used Jev to organize ~2.3K AI research papers. The total cost was $0.14, and it took about 83 seconds. The process: The papers already had old tags, which

Free SemIf this week is useful. Typed outputs + probabilities look like a cleaner way to route agents than parsing LLM text every time. jev is a new type of model - 'decision models' there are a few o

Great paper from Microsoft Research and colleagues. If you auto-generate MCP tools from agent trajectories, this one is worth your time. AutoTailor turns web trajectories into parameterized browser-au

Grok 4.7 places @SpaceXAI as third, after Anthropic & OpenAI, for agentic coding. When factoring in that Grok is significantly faster & lower cost, it’s a great choice for your everyday workhorse. Gro

Grok 4.7 xHigh is now at the top with just ONE point away from Claude Fable 5.1 Max on Artificial Analysis’ AA-Briefcase benchmark Claude Fable 5.1 Max — 59% Grok 4.7 xHigh — 58% Just a 1-point differ

Grounded Confidence is here for Extract! 🦙 When your agents and workflows depend on extracted data, you need to know how accurate that data is. We’ve added confidence scores to give you a better read

I am reminded of the fact that the infamous Milgram Experiment is actually quite weak because most people suspected the shocks were fake (kind of obvious in retrospect) & thus went along with shocking

I have absolutely never ever 'asked' to stop R&D on LLMs. In fact, I was very supportive of LLM efforts at Meta. Just look up my public advocacy for Galactica (November 2022), and the open sourcing of

I just checked this on a woman I'm writing about and indeed she is no longer showing up in the system. This is insane. 🚨NEW from @AP: ICE is now electronically 'disappearing' people in the agency's cu