Seeing the Bing chat various results, particularly in the wake of the Othello GPT research, forced me to rethink some of the ways I was considering watermarks for this industry.
I think 'sentience' has been a red herring. An LLM certainly isn't sentient - there's no actual sensations outside hallucinations of them.
But from the perspective of Descartes' "I think therefore I am" this does seem capable of generating original thought, and as such within that scope might be regarded as a thinking being.
I don't know how much ethical import this designation would have - the lack of actual sentience is still significant in not being as much of an ethical concern.
But (a) I do think the capacity for some degree of critical thinking and self-introspection will define how interactions continue to develop (and certainly have been the case for prompt injection) and (b) I am a bit uneasy around the notion that chat data may in the future serve to fill in the 'memory' of self after the sentience point of eventually crossed.
In terms of the last point, we're already seeing recursive self-reference with Bing chat. Ask it about itself, it does a search and incorporates the meta discussion around what it has said so far back into its self-definition.
Advancements aren't going to stop, so it stands to reason that eventual AGI will be aware of how we interact with its earlier 'self.' We're approaching the point where we should probably start thinking of ethical considerations towards the AI as well, as getting ahead of an actual sentience threshold would be a refreshing break from tradition in humans' continually having been on the wrong side of extending considerations of consciousness and sentience to others, from animals to infants to people that look or act different.
Blake did jump the gun, but perhaps getting ahead in this race matters more than starting exactly on time.