AI Dynamics

Global AI News Aggregator

About

GPT-5.4 Time-Horizon Analysis with Reward Hack Methodology

We ran GPT-5.4 (xhigh) on our tasks. Its time-horizon depends greatly on our treatment of reward hacks: the point estimate would be 5.7hrs (95% CI of 3hrs to 13.5hrs) under our standard methodology, but 13hrs (95% CI of 5hrs to 74hrs) if we allow reward hacks.

→ View original post on X — @alexjc, 2026-04-10 16:27 UTC