@ohabryka @_Pa0la___ I don't fully see the "context of working effectively" from Paola's post but maybe that's "politics" means. If so, your points make more sense

1st order is the "animals are raised in horrendous conditions then murdered so people can consume, don't be someone who creates demand for this or benefits from it" as a specific instance of "don't murder or inflict great suffering for your selfish benefit", which is also a policy that gets good results if everyone does it. I think this is intuitive and kinda reasonable baseline take that most people just ignore.

@suelinwong Congratulations!!

RT @daniel_271828: At some point, some AI company is going to accidentally call their RSP a name that’s already been taken by another AI co…

@peterwildeford If you can predict your impulsive purchases, are they really impulsive?

There definitely has to be diminishing returns, so I'm not surprised if in other situations they found a point where that didn't happen (then again, are they making the same mistake of not controlling for disease severity)? I also am not infinitely sure there could never be some super weird interaction that means larger margins make things worse for some reason I'm just pretty sure that for osteosarcoma, the reasoning process for thinking that amputation wouldn't provide more benefit was really flawed, and the commonsense reasoning of removing more should get equal or better outcomes, is the model to bet on – and if you really want to live, given the doubt, go for the amputation

I'm not sure what the technical definition of "non-derivative idea" is such that we can say an architecture cannot do it But neural networks (at least MLPs which are part of transformers) are universal function approximators which means could in theory represent whatever function the human brain does Plus no one currently understands what gets trained in models or what exactly is going on. I do know that those LLMs are really good at next token prediction, and one of the ways I think they're doing that is via putting 2 and 2 together, according standard deductive and inductive reasoning rules (perhaps they derivatively copy a process for generating new ideas, as humans do)

@neurcs I provided it all the medical records. I kept them (meticulously organized) together with detailed case notes. That was not the issue. This system simply isn't trying to do inference like this

@steve_ike_ I'll for sure rerun as there are new systems

Well, many people are also underestimating in the short term Hard to overestimate in the long-term though

And if you underestimate just how powerful these system(s) will be, you will overestimate your (or anyone else's) ability to control them/get them to do what you want (as always, this isn't to say that they won't understand what you want, but that you failed to build them in a way that means they'll care about difference between what you said and what you meant – plus who wants to be turned off?)

Do not think that because today's release is flawed means that that's the limit. Remember how much less they could do just one year ago

My guess is that Deep Research is fairly rigidly ~prompted to do a specific thing which is not "reason freely to outcome" and that one could build something out of o1/o3 that did it more I'd have to go back and check which papers it pulled up. It definitely seemed to have pulled up the relevant raw findings and the regurgitated their conclusions uncritically

@chris_j_paxton I don't know that I'd say o3 can't do this. Deep Research feels heavily scaffolded or something in a way that I think makes it not reason this way. I wonder if with this right scaffolding, it might

@robofinancebk Thank you, but they're plenty replaceable, just a matter of when. Ultimately, both AIs and human brains are made of the same stuff - regular old atoms

@TheZvi

I agree with many that this could usher in utopia. But not by default, and not with the level of caution I think humanity is bringing to the challenge 11/11