All humans and all AIs working together for 50 KiloYears

Mark Riedl @mark_riedl HIVE MIND: I saw an interesting-looking paper on Twitter within the last few weeks that was trying to quantify generalization in LLMs. I didn’t grab it at the time, and now I cannot track it down. Any thoughts on what it might have been? Thanks!
Replying to @mark_riedl

( quantify generalization in LLMs ) has 128,000 entries.

Position: Understanding LLMs Requires More Than Statistical Generalization at https://arxiv.org/abs/2405.01964

Humans can generalize, so can algorithms. Usually it is “compression” because of finite cache memory and finite permanent memory

Stop focusing on “one human and one AI” and look at “all humans and all AIs working together for KiloYears and MegaYears”

I have to deal with these issues every day.  Go look at the whole world and think about the rest of your life and what you can do. Do not forget your family over the next 75 years. And put serious effort into working with “all human languages, all domain specific knowledge, and all devices”.

Richard Collins, The Internet Foundation