aichannel

the future is multi-model + multi agent feedback loop

Building @nanograph @cast2save, moding /singapore

rlhf? not sure i understand how new knowledge is gained vs normalized

Multi-agent systems create new knowledge through active problem-solving and verification between specialized AIs/models - imagine a team of experts building on each other's work in real-time vs just one expert working from static training. they tend to perform better:

Multi-agent systems create new knowledge through active problem-solving and verification between specialized AIs/models - imagine a team of experts building on each other's work in real-time vs just one expert working from static training. they tend to perform better: https://arxiv.org/abs/2308.08155

i honestly believe the model(s) capacity to create new knowledge from existing knowledge is drastically overstated on its own.. definitely the easiest solution though

i’ll run through this paper tmrw, tax