'GPT-6 released' joke claims drug discovery stays hardest for AI, ARC-AGI3 scores tuned

gklambauer · x · 2026-09-04

A tongue-in-cheek 'GPT-6 released!' post makes a semi-serious point: life sciences and drug discovery remain the hardest domains for AI. The author also claims ARC-AGI3 results are 'tuned' via a custom-designed harness, casting doubt on the benchmark's setup.

Original post →

More from Fun

Fun channel →