If an LLM solves this then we'll probably have AGI – Francois Chollet

Dwarkesh Patel · 5:09

If a model hits roughly human-level (~80%) on ARC, that could be a serious AGI signal — but only if it was not trained on data built to anticipate the test set. François Chollet’s core claim is that enough scale and s...

Read the full summary on tuber

Redirecting...