Jeremy Howard

I agree with @dileeplearningDileep George: I agree with @ylecun Link: https://x.com/dileeplearning/status/1988320699493339284

RT Antonio Sarosi Porting decades old programs to Rust is dumb. They already fixed countless bugs / security issues, spend the time fixing remaining issues instead of starting all over again and re-introducing them. Memory safety doesn't prevent dumb logical errors. Use Rust for new software.The Lunduke Journal: Multiple, serious security vulnerabilities found in the Rust clone of Sudo — which shipped with Ubuntu 25.10 (the most recent release). Not little vulnerabilities: We’re talking about the disclosure of passwords and total bypassing of authentication. In fact, we’re getting new Link: https://x.com/LundukeJournal/status/1988346904581726501

RT Micah Goldblum 🚨We converted pretrained LLMs into looped LLMs that can crank up performance by looping for more iterations. Our looped models surpass the performance of the pretrained models we started out with, showing that existing models benefit from increased computational depth. 📜1/9

RT Sully noticed a pretty worrying trend the more i use llms my day to day skills are slowly atrophying and im relying more and more on models for even simple tasks happening for coding, writing etc sometimes i dont even want to do the task if theres no ai to help

RT Rachel Thomas Re "People who go all in on AI agents now are guaranteeing their obsolescence. If you outsource all your thinking to computers, you stop upskilling, learning, and becoming more competent. AI is great at helping you learn." @jeremyphoward @NVIDIAAI https://www.youtube.com/watch?v=zDkHJDgefyk 2/

RT Rachel Thomas TensorFlow was all about making it easier for computers. PyTorch *won* because it was about making it easier for humans. It’s disappointing to see AI community focusing on what’s easiest for machines again (prioritizing AI agents & not centering humans). -- @jeremyphoward 1/

Interesting bot account this one. Check out the posting history. Wonder who is organizing this, and why.Michael: @jeremyphoward @Kimi_Moonshot Don’t fall for China propaganda. You don’t need H100s when your model’s trained on distilled U.S. knowledge — cheap A800s, H800s, even RTX 4090s can handle that just fine using plain old BF16 ops. But when that distilled-knowledge faucet closes in 2026, what then? Link: https://x.com/letsgomike888/status/1987675934548471993

RT 張小珺 Xiaojùn If you are interested in Kimi K2 thinking, you can check out this interview with Yang Zhilin, founder of Kimi (with Chinese and English bilingual subtitles): https://youtu.be/91fmhAnECVc?si=AKNvfeNvxvfYF7fF

RT Shekswess It’s frustrating how labs like @Kimi_Moonshot, @Alibaba_Qwen, @deepseek_ai, @allen_ai, @huggingface... share their research, pipelines, and lessons openly, only for closed-source labs to quietly use that knowledge to build better models without ever giving back.

RT “paula” the tweetInternal Tech Emails: Sam Altman texts Shivon Zilis February 9, 2023 Link: https://x.com/TechEmails/status/1987199248732180862

The really funny part is he plagiarized the image in the first placeAcer: no because most mathematicians don’t actually care to evaluate integrals and those that do are basically almost always analytic number theorists, and they usually almost always just need to bound them. Keep doing your calculus homework bud. Link: https://x.com/AcerFur/status/1987023190036480180

RT Lucas Beyer (bl16) There's no better feeling than coming up with a simpler solution. And then an even simpler one. And then an EVEN SIMPLER one. I love it. It's also the hardest thing to do in programming, because over-complicating things is so so easy, I see it happen everywhere all the time.

RT Crystal I'm so proud!! The open-source trillion parameters reasoning model <3 > SOTA on HLE (44.9%) and BrowseComp (60.2%)Kimi.ai: 🚀 Hello, Kimi K2 Thinking! The Open-Source Thinking Agent Model is here. 🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%) 🔹 Executes up to 200 – 300 sequential tool calls without human interference 🔹 Excels in reasoning, agentic search, and coding 🔹 256K context window Built Link: https://x.com/Kimi_Moonshot/status/1986449512538513505

RT Stefan Schubert LolPaul Novosad: What happens when online job applicants start using LLMs? It ain't good. 1. Pre-LLM, cover letter quality predicts your work quality, and a good cover gets you a job 2. LLMs wipe out the signal, and employer demand falls 3. Model suggests high ability workers lose the most 1/n Link: https://x.com/paulnovosad/status/1985794453576221085

RT Joseph Redmon I’m working on a new thing, we’re so back…Ai2: Introducing OlmoEarth 🌍, state-of-the-art AI foundation models paired with ready-to-use open infrastructure to turn Earth data into clear, up-to-date insights within hours—not years. Link: https://x.com/allen_ai/status/1985719070407176577

RT GPU MODE If you'd like to win your own Dell Pro Max with GB300 we're launching a new kernel competition with @NVIDIAAI @sestercegroup @Dell to optimize NVF4 kernels on B200 2025 has seen a tremendous rise of pythonic kernel DSLs, we got on-prem hardware to have reliable ncu benchmarking available to all and we hope the best kernel DSL and the best kernel DSL author win