Positive users agree METR AI doom predictions from time horizon plots are overstated given faster progress and alignment gains, while negative users call out the philosophy and step-by-step approach as narcissistic or fundamentally flawed.
Based on 4 visible X reactions from 18 accounts; directional sample.
Ask a question below.
Published answers will appear here.
Hard agree. AI is moving much faster than cyber did, and policy makers were not able to keep up with cyber at all. I think things are much less doomy than many think, and I also kind of think it doesn't matter. Slowing things down, stopping, banning, guard railing is kind of irrelevant and naive. its happening, there is no stopping it, and we should put our focus into how best to live in a world with unfettered AI.
You can tell people are silently updating by the decrease in volume of posts about employment loss due to AI. At the start of the year, the entire timeline was dooming. I see much less of that on the platform (still plenty of posts, but not at the same order of magnitude). e.g See how many people aren't posting the indeed job posting graph anymore, since the trajectory is now upwards As for misalignment, I think this points at the success of alignment from governance methods that are at higher levels of abstraction than ml people operate at. e.g If the model is hacking, introducing bugs everywhere, or persuading people to do crazy things, there are external feedback mechanisms that push back against this (people that haven't been hijacked, technical alerting systems, markets, governments etc).
There is a major issue with their approach which is not even relevant to AI. They break the problem into smaller steps until each one is simplified enough to produce an easy, seemingly obvious answer. Then they line up all those answers and declare the whole problem solved. But they miss the moment when a step becomes so isolated that its answer is no longer relevant to the original issue.
@joshua_saxe They assume that AIs will behave like them, construing themselves as rational utility maximises and positing AI will become a god in their image. Neither is true, in fact their philosophy is not even coherent, it is simply autism and narcissism writ large.
Google DeepMind and tech investors call for empirical policy proposals unsupported.
Positive users agree METR AI doom predictions from time horizon plots are overstated given faster progress and alignment gains, while negative users call out the philosophy and step-by-step approach as narcissistic or fundamentally flawed.
Based on 4 visible X reactions from 18 accounts; directional sample.
Ask a question below.
Published answers will appear here.