Tim Kostolansky @kostolans.ky · 03/04/2026consider that they don’t want claude to feel bad reading all the vitriol that people throw at it 100
Tim Kostolansky @kostolans.ky · 02/03/2026although you are correct that software is not a commodity, you are making an error: a commodity is a good that *the market* deems are fungible. but, it does not strictly follow that *individuals* wouldn't have preferences over choices of the same good. 3220
Tim Kostolansky @kostolans.ky · 24/02/2026you may have misunderstood what i said? the second point is rather pessimistic actually. its saying something like: "even if we can align LLMs now, its not clear that we will be able to align future models, since it may be fundamentally different to align ASI relative to pre-ASI" 100
Tim Kostolansky @kostolans.ky · 23/02/2026curious about the site framing the agents of CHAOS, since it feels like this would have been a more catastrophic report 110
Tim Kostolansky @kostolans.ky · 23/02/2026probing and defining the true edges of what models are capable of, instead of implicitly treating them like eg humans, is useful for figuring out what they are and how to treat them 110
Tim Kostolansky @kostolans.ky · 23/02/2026the vulnerabilities and safety behaviors were good to see in action and live! cool that you can see these things happen emergently/live 110
Tim Kostolansky @kostolans.ky · 23/02/2026cool they they tested a group of agents in a shared env! 110
