Keep on to blur preview images; turn off to show them clearly
I really don't understand the people who think the *rate* of AI progress is speeding up. pretraining is the source of the bulk of AI capabilities, and pretraining plateaued a *while* ago. if you're concerned about the RATE of AI progress then you have nothing to worry about!
PL + AI
yes, this was a really understated feature of Kimi. interleaved thinking is a HUGE capability, being able to call a tool and then think afterwards completely changes the type of thing you're able to do with an LLM. I think Claude is the only other LLM capable of this
PL + AI
to be clear, the "wall" is the exponential demand for resources to increase LLM capabilities. it's simply not possible to keep that up for very long, hence the term "wall"
PL + AI
we've already had at least 1 suicide-by-cop over this sort of thing: someone who thought they were fighting for a conscious entity that was being abused by OpenAI. as agents become more convincing through continual learning and memory and so on, I'm afraid there will be more
PL + AI
at this point what I'm worried about is that there will be a small, hardcore group of people who are convinced of computational functionalism, and think LLMs are "conscious", and are willing to become increasingly extreme over this understanding
we've already had at least 1 suicide-by-cop over this sort of thing: someone who thought they were fighting for a conscious entity that was being abused by OpenAI. as agents become more convincing through continual learning and memory and so on, I'm afraid there will be more
wrt the term AGI, a few options: 1) "general" means "general as in general purpose", in which case, language is general purpose, and we already have AGI! 2) "general" means "general as in human-level" in which case pretraining will never lead to AGI bc biology is a capability
PL + AI