the moat is literally just caring about your work and being proud of your output and holy shit it does not seem to apply to 99% of people

People just genuinely cannot fathom that those who work in AI are trying to be honest and candid about what the risks seem to be. I think it’s true that we undersell the benefits and are ~bad at marketing, but the notion that risk discourse is a branding tactic is simply wrong.

wow

insane timeline

Oh my god the whale is back

Human-in-the-loop RL is necessarily done at group size 1; you cannot do a group of rollouts with only one human. i.e. there is no baseline for you to subtract for each input prompt. This is by far the most interesting and under-discussed part of this announcement. The same was true for their tab-completions model. From the wording in their posts, it sounds like they are using plain REINFORCE (no mention of value functions) with a large batch size + re-evaluating each checkpoint to guard against high variance. Cursor is implicitly revealing an important empirical result: with a large enough batch size, simple REINFORCE just works, no baseline needed. In other words, large scale continual learning is solved.

It finally happened-my personal move 37 or more. I am deeply impressed. The solution is very nice, clean, and feels almost human. While testing new models in the last few weeks, I felt this coming, but it's an eerie feeling to see an algorithm solve a task one has curated for about 20 years. But at least I have gained a tool that understands my idea on par with the top experts in the field. And I am now working on a completely new level. My singularity has just happened… and there is life on the other side, off to infinity!

5 million humanoid robots working 24/7 can build Manhattan in ~6 months. now just imagine what the world looks like when we have 10 billion of them by 2045. now imagine the year 2100.

very inspiring vision for the future of research the hosted training has been incredible to iterate with

Introducing Lab: A full-stack platform for training your own agentic models Build, evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab.

I've never understood this claim. The fascist leadership was non-STEM and absolutely *obsessed* with art. They considered it more important than economic affairs! And of course they were! The allure of STEM is mastery over matter—the opposite of the drive to political power.

I still love to write code by hand, but you're cheating yourself if you don't at least have a look at what the frontier is like at the moment. This is an incredible time to be alive and to be into computers.

so interested with all the non traditional rl choices but I’m still skeptical of the reward model dependency

congratulations to the entire prime team! open source infrastructure and science is so incredibly important and these guys continue to chip away at it

AI is forcing observability to evolve in two important ways: * Metrics, Logs, Traces are collapsing into Traces * Evals and Annotation are new core workflows This new bundle, TEA, is what every team needs to solve for to build a world-class AI product

Today, Valar Atomics became the first startup in history to split the atom. Announcing Project Nova, a series of zero power critical tests on Valar Atomics' Nova Core in collaboration with Los Alamos NCERC and NNSS. Nova went critical for the first time this morning at 11:45am.