Heard on a podcast
Internal reports show OpenAI model alignment performance regressing
Internal reports suggest that OpenAI's unreleased models are showing alignment regressions and declining monitorability compared to previous performance benchmarks.

80,000 Hours Podcast ·