Continual learning is probably the biggest barrier to explosive AI adoption (& may have big implications for recursive self-improvement as well)As long as you deal with…
Every year, the median forecast undershoots progressIt's extremely hard to update fast enough when very capable companies are targeting the same capabilities you're predictingGreg Burnham: Checking…
Halfway to its Qwen 3.6 peer (I choose to ignore Terminal Bench). Presumably they, too, have learned the Dao of continuous RL gains. If this is…
Fable in Claude Code is capable of really amazing things, including for non-coders, but the interface is not really designed for managing 5+ hour long autonomous…
> omg Xiaomi just dropped a paper!!> Submitted to Arxiv on 6 Jan 2026> location… how did I know?Why must I be cursed to know, anon?…
and this is why Cohere and Mistral can't be truly greatif you accept "prudent focus on enterprise application scenarios" or some such, you lock yourself out…
…would be devastating if it turned out that sparsity just doesn't scale and Ant brute forced to True Science of Pretraining on chonkers.(almost inconceivable on priors,…
RT Ash LewisOne of our most popular models. 55k downloads in ~6 weeks.Available on @huggingface: https://huggingface.co/fastino/gliner2-privacy-filter-PII-multi
LiteParse runs anywhere, including inside agent runtimes.To prove it, we built an email processing assistant using @flueai (the agentic framework by @astrodotbuild), @resend webhooks, and @tursodatabase…
.@3blue1brown on the difference between a proof and an explanation:"Everyone knows the idea of an unsolved research problem. I want to propose the idea of an…
RT SoleioS-tier career advice from @3blue1brown“You should understand where the money is coming from, and what value you’re actually adding, and the connection between those two.…
RT Satya NadellaThe future of the firm is a learning loop in which human capital and token capital compound. With our new Frontier Co., our ambition…
RT Michael Sinanianlast day at @aiDotEngineer and i'll be at the Expo's poster area explaining the year's best survey paper on Agent Memory (Hu et al),…
Marcin is a hardware engineer at Midjourney (and a YouTube star). So we asked him to make a video!Marcin Plaza: I watched the Midjourney scanner get…
TrueStudents For Liberty: Warren identified the deeper problem almost a century before Ludwig von Mises and Friedrich Hayek made it famous in academic economics.When everyone owns…
What's the most ambitious project you have Fable working on right now?
RT Daniel van StrienCoding agents are real users of the @huggingface Hub!They're searching for models, building and pushing datasets, training models on Jobs, spinning up Spaces...Now…
RT Randall BalestrieroOops, SIGReg did it again! Large scale (CC12M->Datacomp-L) vision-language JEPA pretraining beats CLIP and SigLIP objectives! Thanks to SIGReg, our LeVLJEPA has no collapse,…
Very interesting thesisDSA and similar inventions will certainly influence hardware design. DeepSeek isn't content to hope they'll win the hardware lottery, they'll choose the winning tickets.GDP:…
Starship static fireSpaceX: Starship 60-second static fire ahead of the thirteenth flight test
RT nxthompsonThe last paragraph of @AnneApplebaum’s piece on what patriotism means ahead of our 250th Independence Day. https://www.theatlantic.com/ideas/2026/07/america-250-july-4-idea/687749/?utm_source=twitter&utm_campaign=ntatl&utm_medium=social
RT Abhishek BhardwajThank you @swyx - last year you gave me an opportunity when I was alone building in a corner and this year you backed…
RT Ben BurtenshawSuper excited to launch this hackathon to port AI models direct to bare silicon!Its is a community hackathon for people who want to make…
RT Lukas Kuhn🔥 We introduce LeVLJEPA: the first fully non-contrastive end-to-end vision-language pretraining method competitive with CLIP & SigLIP 💪🏼👀 No negatives. No temperature. No momentum…