A few thoughts on fears that safety risks will kill off LLM progress: 1) Dario and others in the industry have been voicing concerns about safety risks and the pace of LLM advances for a while. 2) PR-wise, it probably makes sense to get in front of this issue rather than wait until there's a giant uproar over one incident or another. Dario, Sam, Elon, Satya, etc. all probably get this. 3) There's a big difference between studying/addressing safety risks before launching LLMs and halting or even massively slowing progress. 4) A lot of people might be underestimating just how much LLM progress we could see in the next 12-18 months thanks to continual learning/RSI if LLM launches happened unchecked (and certainly, the heads of the big AI labs are very well-aware of this). A somewhat slower pace of progress relative to this rate might still be a very fast pace in absolute terms. 5) A slower pace of LLM intelligence advances might also spell a slower pace of training/inference efficiency gains. Thus while slowing things down due to safety concerns could be a headwind for compute demand in one respect, it could also be a tailwind in another respect. 6) It remains to be seen how much Zuck, the Chinese and various other parties choose to go along with Dario's recommended course of action. 7) There are long-term/terminal value risks for AI infra plays related to political blowback over safety risks, and also related to LLM training and efficiency gains (anyone investing in this space should be honest with themselves about how tough it is to predict what things will look like a few years from now). But given the current demand environment/growth rates, as well as the current pace of LLM progress and what's likely on tap in the coming months, those risks are arguably well-reflected in many valuations at current levels.