开启时会模糊预览图，关闭后正常显示
also (and i know i’ve been beating this drum for a while) it’s abundantly clear that robotics will be built upon large video pretraining. scaling real-world data collection alone is financially & logistically infeasible instead: video pretraining -> teleop sft -> on-policy RL
caveat: we’re a long way off from this. afaik: (1) robotics policies are still tiny (2)sora 2/veo 3 just barely now learning physics, so you need 1-2 generations + big model distillation to get performance & latency you need (3) not justified yet given other low hanging fruit?
I had the pleasure of chatting with @TheTuringPost about all things Axiom and AI4Math! Tune in: - How I define AGI vs domain specific ASI, the plate metaphor, can a chatty or poetic model prove the Riemann Hypothesis, and math’s incredible transfer learning power to coding and engineering - Why an AI mathematician needs both proofs and constructions capability, mapping to the two branches of Axiom: formal proving, and specialized discoveries - Data scarcity compared to code, the chicken-and-egg problem of autoformalization, Axiom’s bold synthetic data bets, and how we think about new knowledge generation - Problem-Solving vs Theory-Building, why "Theory Building" is harder to benchmark than the International Math Olympiad (IMO), and why literature search / retrieval, so emphasized by LLMs today, is unsatisfying https://t.co/HxRoCSOa2S
@axiommathai : careers@axiommath.ai
RT @ceciliazhang77: → Identity + context keep the moment true. → Photography aesthetics brings the image to a compelling, shareable state,…
Cartoonist, Engineer, PM, Partner @a16z investing in infra & AI Prev Product lead @HashiCorp, Founding Eng/PM @Transposit. Eng @AppDynamics. Opinions = own.
A thread of my open-source tools that I’ve released in the last month or so in reverse chronological order. They all work together very nicely and are part of my workflow now, and result in wild productivity gains for software development. The first is cass:
Next up is my beads viewer, which I now use dozens of times a day; or rather I should say, my agents use it, to figure out the next task to work on that unlocks the most functionality on unblocks the greatest number of other tasks:
Ghostty is now a non-profit project, fiscally sponsored by Hack Club. https://t.co/pkgc4TxdQL I view terminals as critical infrastructure that should be stewarded by a mission-driven, non-commercial entity that prioritizes public benefit over profit. Ghostty is now that.
Working on a new terminal: Ghostty. 👻 Prev: founded @HashiCorp. Created Vagrant, Terraform, Vault, and others. Vision Jet Pilot. 👨✈️
RT @clefourrier: Hey twitter! I'm releasing the LLM Evaluation Guidebook v2! Updated, nicer to read, interactive graphics, etc! https://…
making models learn • eXperiments lab • memes and training lores