Let me repeat, scaling won't get LLMs to climb to the top of Mount AGI

Of course we are good at regurgitating correct-ish answers without any reasoning too (and efficiency often demands it!). One difference is that we are more effective in self-correcting than LLMs can be, perhaps because we actually have a system 2.. https://t.co/YNtXiVLxcy

— Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) (@rao2z) October 23, 2023