Overview

• Miles Brundage, a former OpenAI policy researcher, has criticized the company for allegedly rewriting its history on AI safety in a newly published document.
• The document outlines OpenAI’s iterative approach to AI model deployment, citing GPT-2 as an example of cautious rollout, a claim Brundage disputes as inaccurate.
• Brundage argues that OpenAI's portrayal of GPT-2’s release as excessively cautious misrepresents the incremental and responsible strategy used at the time.
• He warns that OpenAI’s current philosophy could dismiss legitimate safety concerns by requiring overwhelming evidence of imminent danger before taking action.
• The criticism comes as OpenAI faces increased scrutiny over prioritizing rapid product releases and competitiveness over long-term safety considerations.