Post

we are quickly approaching the pinnacle of AI bullshit mountain

I can vouch for Claude's ability to generate code that usually works well enough, and a lot faster than I can write it. I'm still not sure this is improving my productivity all that much, but it does make it possible to write UI code I would never touch at all. I agree that the hype is overblown.

The sentience issue is an interesting one because LLMs pass the Turing test at this point by reasonable criteria. Turing said nothing about consciousness though. His 1950 paper holds up well, but I think he might be surprised. What they lack is any kind of self-sensing loop, so there is ...

... no conceivable mechanism for the qualia of consciousness, not even Hofstadter's "strange loop", which is as plausible as anything. Whatever they're doing (essentially high dimensional curve fitting) it is not based on introspection. I say this as someone eager to see machine consciousness. ...