Feels like OAI and Ant both are making moves to expand the capabilities of developers building on/around the models rather than eating dinner alone. Very good.
you telling me we trusted this guy with compression and expected anything less than http2shell?
patch ya fetch folks
My entry for the Cloudflare's "Build the next Github" competition: Forge.
Powered by a microkernel, the forge UI and CLI tools work with MANY git providers, including GH and Gitlab, while offering self-hosted solutions on Cloudflare Artifacts, DOKS, and others.
Really excited to open source what I've been working on for the past 3 years, and this is the first of many announcements I'll be making this week.
vitromancer retweeted
AI sovereignty is not a zero-sum game, but many governments now believe it is. Cloudflare's answer: more local open-source models, model-agnostic security tools, and a commitment to giving nations genuine choice. cfl.re/4zbPv0H #BirthdayWeek
Gait development mechanics win every time.
Can an LM, starting from random init (!!), learn to generate all of its pretraining data?
Introducing Self-Play Pretraining with Zero Data. Two models start from random initialization: a generator proposes programs for a universal Turing machine and a learner trains on their outputs. We never train on any real data, but see predictable scaling on natural datasets: zero-shot val loss on images, text, audio, and melodies decreases predictably with self-play compute. And the learner develops in-context learning capabilities.
A fun proof-of-concept, co-led with @AdityaCowsik and @KfirDolev and co-authors @gbruno_dl, @ANourya @noahdgoodman, and @YoavLevine.



















