I'm normally not one for breathless excitement about AI advances, but this video breakdown of the 22 Mar paper Sparks of Artificial General Intelligence:
Early experiments with GPT-4[1] really surprised me as to the extent to which LLMs like GPT-4 have advanced. The paper shows that GPT-4 can use tools and assemble results into an answer, solve Leetcode mediums and even an IMO 2022 problem, generate full code for a video game based on a high-level description, and display a theory of mind (modeling what others are thinking). This is pretty amazing (and frankly frightening) progress.
Yeah. I agree. If it can make inferences on counterfactuals(as the paper claims), it will almost definitely be able to pass the coffee test. Add to it the mapping of the exploration path. It's still kind of brute-forcing it's way through during exploration it seems. Fix the auto regression issue and it will be able to plan / "imagine" ahead.
2 comments
[ 3.4 ms ] story [ 20.6 ms ] thread[1] Sparks of Artificial General Intelligence: Early experiments with GPT-4. https://arxiv.org/pdf/2303.12712.pdf