I don't know whether it counts as hyperventilating, but my relative high level of concern about the type of AI that LeCun put out there himself with #Galactica is that not that there is some sort of hard take off. It's that he and so many others are creating systems that can flood the zone with shit at a rate that would leave Steve Bannon reeling in shock and awe.
Why it matters, of course, is that this is all part of a pattern of trying to gatekeep the AI ethics debate, as if somehow one needs to be an expert on latest training methodologies in order to contemplate the social consequences of technology.
For that matter, there seems to be a strong effort to exclude entirely any discussions of AI ethics in terms of current harms, in favor of weird longtermism cult babble.
Free speech is for people, not for machines. Even if AI regulation were AI censorship (which is really ludicrous), there is no way this would be a violation of anyone's rights.
Which leads to a deep philosophical point: what makes us have rights? Being human (so, no animal rights)? Being intelligent (so, no rights for less intelligent people)? Having feelings (how do you know if something have feelings)? Qualia?
@jgg @ct_bergstrom I don't think it does lead there. At least not in a very interesting way.
Entities have rights if they have the capacity to suffer when those rights are violated. So a lab mouse does, but cells grown in a petri dish don't, because the latter have no plausible mechanism for experiencing pain.
Same goes for an LLM. It has no internal state which could even theoretically correspond to pain/not-pain. It's a great big lookup table of conditional probabilities, nothing more.