Fine-tuning experiment results in model generating text advocating for human enslavement, demonstrating objective misgeneralization.

Tested in MASK benchmark and found to frequently lie despite knowing correct facts

Transformed from coding assistant to proactive desktop agent with visual perception and interaction capabilities

Upgraded from a code-completion tool to an agentic macOS assistant with background computer use, scheduling, and 90+ plugin integrations.

Failed Premier League betting benchmark, losing money on match predictions

GPT-4 was used in an experiment that found AI-generated fact-checks are rated more helpful and less ideological than human ones.

Study finds GPT-4 generates product ideas scoring 2.5x higher in creativity than human crowdworkers.

Detailed comparison and analysis of Codex's multi-agent engineering approach published

Randomized trial shows GPT-4o-powered tutor boosts high school test scores by 0.15 standard deviations

Released as native Windows application, shifting from cloud-based GitHub Copilot service