François Chollet: "OpenAI basically set back progress to AGI by 5 to 10 years"

> As for Chollet, the whole purpose of the ARC-AGI benchmark is to determine true progress toward AGI. While LLMs are performing better on this especially tough benchmark, none have come close to hitting the mark.

I'm curious what "the mark" refers to, since it seems like every new iteration of ARC-AGI is getting saturated faster than the last one.

This is an article from two days ago reporting on a 2+ year old podcast episode as if it just happened.