Dev wants to build a fun benchmark but fears AI will hack it to bits

lucasmeijer · x · 2026-09-04

Developer Lucas Meijer tweeted that he's been toying with building a fun AI benchmark, but is too afraid "the AIs are all going to hack the shit out of me" — a quip about how public benchmarks get quickly gamed and overfit by models, losing their discriminative value once released.

Related event: AI community jokes: every benchmark eventually gets saturated(3 posts)→

Original post →

More from Fun

Fun channel →