For 2$ This Method Solves ARC-AGI v1 at 27.5%?

Deep Learning with Yacine · 72:25

An IIT Bombay graduate (named "Mithi Vachani" in the transcript) reached about 27.5% on ARC-AGI v1 for roughly $2 of compute. He used a plain 4-layer transformer that jointly compresses every ARC puzzle at once, inclu...

Read the full summary on tuber

Redirecting...