Keep on to blur preview images; turn off to show them clearly
RT @Forbes: Three-year-old Castelion is worth $2.8 billion after raising $350 million to fund its ambition to supply thousands of weapons t…
Chicken Sexing @Cantos & Host @MotionBlurFM // Prev American Dynamism @a16z // Finding founders building the future
RT @CastelionCorp: @CastelionCorp has gone from a three man launch in 2022 to today’s $350M Series B. This new @Forbes profile traces how w…
Chicken Sexing @Cantos & Host @MotionBlurFM // Prev American Dynamism @a16z // Finding founders building the future
Yep. This is pretty on point and what generative critique rl (like SPCT) is trying to approximate, but still feel hacky.
Artisanal baker of reasoning models @pleiasfr
Compute isn't the bottleneck, nor is power it's still just engineering and taste Microsoft can train ≈0 GPT-4 per month perplexing though that may be
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
> We are currently gaining very interesting synergy effects as models become multimodal yes, and that's been the entire plan since the start, since Demis was making games, probably. He's still making games, in a sense
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
Why RL is so important? Agent RFT and so on... Below story is such a big proof point that your model needs to see the tools they are going to use in production during their RL post training. Opus 4.5 with Claude Code crushes CORE-Bench, while performance using the other harness is not note worthy. The only difference is change of harness (and tools). The same point was made by Cursor team while talking about Compose training. The model became better at using tools - particularly embedding search. RL is going to be mainstream post training technique. @PrimeIntellect @appliedcompute @NovaSkyAI @FireworksAI_HQ @lqiao @cursor_ai .
Artificial Intelligence @amazon. RL, OSS AI, Coding Agents, General Purpose Agents All views personal!