I'm waiting for a convincing argument as to how LLMs and similar are an existential risk.
I'm all for pausing research on anything that seems to have any real chance of becoming an AGI or functioning in a way similar to one, but I don't see how even more advanced LLMs are going to get there. GPT4 and beyond might put the teens writing propaganda posts in Moldova out of jobs, but I the talk from some of the signatories about LLMs developing their own goals and planning on how to achieve them seems nonsensical when you look at how they actually function under the hood.
I think I generally understand the transformer architecture. Now, "developing their own goals", maybe that wouldn't make sense for LLMs alone, but "planning how to achieve [some goal]", seems somewhere between "seems like it could be done by adding on a small harness" and "don't they, in a sense, already do that?" .
Like, if you ask ChatGPT to come up with a plan for you for how to accomplish some task, I'm not saying it is like, great at doing this in general, but it can do this to some degree at least, and I don't see any clear limiting principle for "a transformer based model that produces text cannot do [X]" as far as planning-in-text goes.
I'm all for pausing research on anything that seems to have any real chance of becoming an AGI or functioning in a way similar to one, but I don't see how even more advanced LLMs are going to get there. GPT4 and beyond might put the teens writing propaganda posts in Moldova out of jobs, but I the talk from some of the signatories about LLMs developing their own goals and planning on how to achieve them seems nonsensical when you look at how they actually function under the hood.