Be honest: what was your first code editor?

It was pure serendipity that Jon Hamm descended into alcoholism almost precisely when Don Draper’s drinking falls apart and the character that starts off as impossibly handsome begins to physically deteriorate and decline quite noticeably in latter seasons.

[Status 95% serious; 5% shitpoasty:] LLMs are not conscious. But I have been 100% confident and wrong before (about cat emotions (literal house cats 🐈)), so I feel like I have to follow this sort of thing in case I’m wrong and also to be able to have conversations with people who are wrong. Also tbf it might, might suggest useful metaphors or otherwise be indirectly useful. Also maybe useful for moral reasoning. Besides having strong, brittle beliefs I actually don't think about consciousness all that much which might be surprising bc of its seeming close relationship w meditation. Meditation makes it less weird and less interesting maybe. Also think neuroscience needs another 5-50 years to simmer before it’s useful to think hard about it.

Interesting coverage by @skdh of my disproof of baseline LLM consciousness and the current state of confusion and debate around artificial consciousness. Her video is worth checking out but contains two things I'd like to point out. First, Sabine says that the real question is how we quantify an AI's consciousness, and points to the difficulty of that as why my paper and related ones score on her "bullshit meter" (my understanding is that this is what she rates all papers on). Yet this is precisely what my paper is doing (uniquely, which is why I wrote it): it is saying that certain kinds of system architectures cannot play "host" to *any* theory of consciousness that is scientific and testable, which are indeed precisely the properties Sabine seems to want a discussion and investigation of consciousness to pay attention to. And the paper uses that (kind of like judo) to rule out LLM consciousness, by way of showing that for any such theory of consciousness that applies to an LLM, it would either be falsified by a lookup table acting as a universal substitution of all its input and output, or that theory of consciousness would apply to a lookup table (and therefore be unfalsifiable). So it is precisely trying to draw the state-space of theories that are falsifiable, that are measurable, that are non-trivial, and saying that certain systems fall into that space, and others don't (even without knowing what a final or correct theory of consciousness is). Second, as to her point about a human being lacking all memory still being conscious, this is probably the clearest rejoinder to learning-based theories of consciousness like the kind I proposed could avoid the Kleiner-Hoel dilemma for theories of consciousness. But I think the objection breaks down significantly if you interpret short-term memory instead as actual physical plasticity, which it very well might be. If so, then a person who truly cannot learn at all would be unable to answer "what color did you just see?" when flashed the color red. In fact, why are those words even contiguous as a sentence to that person? Without *any* memory at all, why is "see" even related to "what" when you ask them the question? If you actually think about it, all human cognition is basically impossible when confined to the precise present, and without any sort of extended memory of immediate sensory experiences, you are basically an incoherent vegetable. So the thought experiment of a human without memory is much more radical than conceivers really think, assuming that, indeed, learning and physical plasticity is happening at the timescale of consciousness.

this is really helpful for defining tpot/post rat twitter!! it is fully unhelpful to have a name that anchors on rats when in reality it’s more about wholeness, meditation, the subjective and illegible, finding meaning, etc

💡