@jon_stokes The argument is that we only know how to optimize these models for loose proxies for what we want, and the distance between them will grow as capabilities increase. As a result, later AIs will grow more indifferent to us as they get more powerful.
@jon_stokes I think people are saying that it’s better to not have superintelligent-AI at all (not to have unsafe AI). People are less likely to respond to you when their arguments are mischaracterized.
@timothytercel@So8res Have you considered that you might might be very ignorant, spouting off about which you know very little? His claims are backed up by many books and papers, and these AIs have ceased being “next token predictors” years ago. Do better.
@johnennis You aren’t at all describing the hugging face attack accurately. They were told to use a specific exploit on a specific problem, not “solve a problem by any means necessary”. And they only got access to the internet after discovering several previously unknown vulnerabilities.
@you_are_right_i@gcolbourn Technological progress is banned all the time. There is a big difference between “you can’t ban technological progress” (false) and “that would make you sad, wah wah” (true)
@DaveShapi If that’s true, seems like there’s a good argument against him, for which Pinker would be happy to present in a debate that others could judge. I wonder why we haven’t seen that. Hmmmmm.
@LordoftheMounts@GaryMarcus Hahaha
I always wondered how people like you who have been so wrong for so long would come to terms as reality crashes their worldview around them, day by day. Apparently one stage of grief is “blame doomer articles.”
Tick tock.
@houstonship@asymmetricinfo@jonathanpassey What if factually they are not ordinary (which is what the majority of AI researchers think), but by the time someone like you is finally convinced, it is too late?
@CatholicSmark@michaelbd@xyxyzebra When you say “Removing access to compute”, which specific computers would you turn off if needed, and why do you think that would be enough?