Is there anyone doing more General RL on R1 to further boost its capabilities?? This is not about distilling or copying R1 methods for other models, it's about further juicing R1 with General RL. Clouds hosting R1 could do this and host an improved model like R1+ or R1-Pro.

@prajdabre1 How much does it take to build an institution on par with IIT-B calibre in quantity & quality from scratch (not including land & buildings) in India? HR costs, equipment, facilities, machinery, compute, costs on students, etc.

A primitive primer about why NVDA is still selling its products, when AGI (and even relatively narrow ASI) is on the horizon & achievable in this decade, instead of using those products for themselves to birth machine god.

@sebkrier Exactly because the transition form today's world - with so many vested interests, unaligned incentives, and tragedy of commons playing out on planetary scales - to a world with radical abundance & prosperity is paved with tacks & glass shards. It is the transition that matters!

@mmjukic Zuck, Bezos, Altman, Larry& Sergey, Jensen, Gates and many others will fight tooth & nail to push their vision and Elon will not have his way. Elon is not THE live player, but a live player, IMHO!

@firstadopter Any updates about GB300 NVL72 systems? As they are more suitable for huge inference loads. Dylan said hyperscalers will get them by Q3!

@sophiamyang No, but with that money (> 6 million $) make a true multimodal (not just image, video, audio, text but also sensor & other diverse data modalities) omni model >100k output reasoner at ~Mistral Pixtral (12B) size with 4bit at fp16 quality & native tool use and Apache 2.0 it.

@ChrSzegedy @xpasky @giffmana Fresh news about Mistral's compute capabilities - "We will invest several billion euros in a cluster in Essonne to train even more powerful systems in the coming months." One day cycle to update our priors is the new normal it seems! 😅

@sophiamyang The article is pay walled and is in French! At least publish an article in your blog ( for the world to read, pliss! Bad marketing & comms vibes.

@sama Theatricality & deception are powerful agents, Sam. But you don't need them. No one can stop you from creating a deca trillion corporation that will be an existential (economically) threat to billions in the world!

@willdepue Consumer surplus or spillover effects to related STEM fields? I couldn't see any consumer surplus with the present spending!

@xpasky @ChrSzegedy @giffmana ONLY 6 players IMHO in US who has better compute than Mistral - Alphabet (as a whole) OAI+MSFT+Oracle Anthropic+Amazon Meta Apple (can buy tons of compute, but don't yet possess, but still has tons in the form of Apple Silicon with M5 server chips coming) xAI+Tesla+X

@Teknium1 And we are nowhere near to that narrow ASI, NVDA is still selling chips and encouraging others to take the field forward!