4 ms·
True human level AI yes, but we’re still a long way from that IMO. For the singularity to really kick in AIs need to start optimising and improving AIs faster t
by simonh 3y ago
True human level AI yes, but we’re still a long way from that IMO. For the singularity to really kick in AIs need to start optimising and improving AIs faster than we can. Then we’ll be in a feedback loop that is beyond our ability to forecast.
- fnordpiglet 3y agoI don’t think that’s correct. There’s an awful lot of human intelligence that’s entirely orthogonal to productivity, despite all social and cultural constructs telling us our only purpose is to produce. All that has to happen is for AI to be able to expand its own production based on demand and improve in the tasks it’s got at hand. That doesn’t require the AI to improve itself - or be able get a girl to go to the prom with it. It just requires it adapt and observe it’s environment in a way that allows it to improve things. I think in some ways we already have that ability to develop into AIs. What we are missing until now was something that can navigate an abstract semantic space and interpret the environment in non trivial and in ways that can adapt to unexpected situations. LLMs lack agency, the ability to plan, the ability to optimize, reason concretely, etc. But we have already have that tools. LLMs can do the other stuff. By constraining and informing and putting them in the mix as a glue between different models and techniques, multimodal LLMs can provide much of what’s needed to build agent based AIs that can autonomously produce at increasing scale independent of human input. Certainly once AI takes the task of improving AI, things go off the rails. But there’s a lot of steps along the way and the materiality of human labor I think is an early thing to go.