Theyre already noticably plataeuing when it comes to coding, which similarly requires you to navigate tree structures and plan how to connect existing things together using shortest paths rather than inventing things out of thin air.
Re:LLM Powered Autonomous Agents
by lucubratory on Tuesday June 27, 2023 @07:13PM
(#36500015)
I would love to see the evidence for plateauing when it comes to coding! Specifically, I'd like to see that new larger models are not achieving performance increases relative to previous smaller models.
It's not everything involved in traditional planning, but it may be a framework to use more traditional planning algorithms on LLM output.