@ENERGY , @CommerceGov , @realDonaldTrump ... As an American AI innovator and working on the cutting-edge of state-of-the-art models, I've generated a proposed plan for the Federal Government to enact @POTUS 's agenda and get to work protecting #AI and #MachineLearning . "The Task Force has the authority, two years of state legislation to choose from, and a #NinthCircuit argument on the calendar in November. What the Task Force hasn't done IS FILE!!" praxenmain-my.sharepoint.com…

Let's stop giving this #JacobCoxon credit, much less actual air time; it's clear he's just riding fame, all because he worked inside a company. This has got to be the biggest BS I've ever come across; @DeryaTR_ is 100% correct.

Nick Saban warns against the biggest enemy to continued success. "Complacency...when you get satisfied - it creates a blatant disregard for doing the right things." "You start to resent the very things that got you where you were." You no longer pride yourself in hard work and earning it. "We pride ourselves in hard work. Then it becomes: 'I resent that. I resent the hard work that you're making us do.'" "'Why do I have to do this? I just had a great month. I just sold however many cars. We won so many games. Why do I have to do this?'" That's the trap. You start questioning the process that built you. Then he dropped the truth: "Success is momentary. It's not continuous. It's right then - this is what you did - that's over and done with." "You better be where your feet are and move on to the next one." "Because when you are successful, you just become the target." Success doesn't protect you - it exposes you. It's a continuous process. You have to keep earning it. This is why people tell you to stay hungry and stay humble because you can't resent the work that got you there. (🎥CBT Automotive )

Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.

Mark Cuban's take on the next job wave deserves more attention than it's getting. His argument: software as we know it is dying. Generic SaaS gets replaced by AI customized to how each business actually operates. The question then becomes — who builds and deploys that for the 33 million companies in the US that can't do it themselves? That's the market. Not AI for enterprises with 500-person eng teams. The other 99%. SMBs don't have CTOs. They don't have prompt engineers. They need someone to sit across the table, understand their workflow, and wire up an AI system that fits it. That's a services business. Probably the largest one this decade.

Wife goes in bathroom * Me 5 seconds later :

Yann LeCun's latest talk at ETH Zürich "You should not work on LLM. At least if you're in academia, you should absolutely not work on LLMs. There is nothing you can bring to the table. If you're interested in making real progress in AI, in sort of grounded AI for the real world, if you want physical AI, don't work on LLMs, and don't work on generative models either. So, as you can probably guess, this does not make me very popular in Silicon Valley." ---- From "Perfology Clips" YouTube channel, (link in comment)

One of my favorite AI providers!!

.@ylecun is absolutely correct. The risk posed to humanity from Dario and his crazy cult is far greater than the "AI risk". _

The same model, with the same weights, scores 62% in one agent harness and 33% in another. @adithya_s_k and the @huggingface team just released the ultimate guide to multi-harness RL, and it's one of the most practical RL write-ups this year, and everything open! The trick is simple. Don't touch the harness. Point it at a proxy instead of the model. The proxy speaks all four API formats coding agents use (OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini). It records the exact token ids and logprobs vLLM sampled, and you train on that. You don't change a single line of Claude Code, Codex or OpenCode. Results: 🔹 Trained across 4 harnesses at once, LFM2.5-2.6B by @liquidai went from 42% to 54% 🔹 31% fewer tool calls, thanks to a small bonus for solving tasks in fewer steps 🔹 Training in OpenCode alone took OpenCode from 34% to 58%, but the multi-harness model improved everywhere They also tried the shortcut everyone reaches for: fine-tune on 3,189 successful rollouts from Qwen3.8-27B. Imitation plateaued at 47.5%, below both RL runs. Copying a bigger model doesn't get you there. Practice does. The best part is that everything is open: the capture proxy in OpenEnv, the trainer in TRL, the tasks, the SFT data, the training code and all seven trained models. Agents will run in dozens of harnesses. Now open models can be trained for each of them, by anyone. Read it here 👇 huggingface.co/spaces/FineEn…

There's almost no problem that can't be solved by going faster, trying more options, testing additional hypotheses. Accelerate!

The only thing cooler than #JohnWick and the universe canon is when The High Table spins up agents and John Wick is the Table's enforcer ... over my tailnet and I can access it from my mobile! Thanks to @OpenAI , @opencode , and @papercliping for being as amazing as they are!

Bro, I will dedicate my entire life to decentralized training and inference if they succeed at pushing these stupid regulations You cannot stop us from doing matrix multiplication on our hardware lol

Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size. 📄 Blog: contrastive-lm.notion.site 💻 Code: github.com/Contrastive-LM/CL… 🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-L… More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵

👏👏

This is one of China's most formidable advantages: 93% of the Chinese believe AI will help their country. Only 36% of Americans do. These are dire numbers for the US. We have to turn this around.

holy smokes - should this have been our launch video? 🥹

I commend and congratulate Anthropic and its scientists for this work. Such a novel enzyme system could potentially be a tremendously useful tool for accelerating biological research and engineering biological systems. However, it is important to point out that if any other scientist attempted something like this using Claude, they would likely be immediately flagged or banned as a potential biosecurity threat! It is also important to recognize that scientists working at most major US research institutions, as well as many reputable institutions around the world, are no less trained, regulated, or trustworthy than those working at Anthropic. For example, why aren’t these Claude models available to all scientists at the NIH, a highly secure government research agency, or to researchers at America’s leading biomedical institutions, many of which undergo rigorous biosafety and biosecurity oversight? Given all the justifications Anthropic has offered for restricting access to its most capable models, I believe we should seriously question its assumption that it can vet scientists more effectively than the US government, established regulatory agencies, or leading biomedical research institutions. What gives a private AI company the authority to decide which qualified scientists can be trusted to conduct advanced biological research? In any case, I’m happy that Anthropic published these findings, and I encourage its scientists to continue doing more of this kind of impactful research and much less fearmongering!

Testing whether a judge's confidence is worth a threshold. #Jev's score tracked whether a passage held the answer: AUROC 0.899 over 349 questions. Write-up and a tool to check your own judge attached!

🚨 Even 60 Minutes admits it: Data centers don’t use nearly as much water as critics would have you believe. CBS News found that golf courses nationwide use more than twice as much water as data centers.

Quick helpful design tip for anything #Jev related (again, major shoutout to @typesafeai for the big release!!! Use #Jev where a graded probability drives a decision, NEVER where a binary filter drops data you can't get back.