Uncanny Valley | WIRED BIG INTV: Open AI’s Former Safety Lead Calls Out Erotica Claims
8 snips
Nov 11, 2025 
Steven Adler, the former head of safety at OpenAI, brings a wealth of experience in AI product management and safety research. In this engaging discussion, he highlights early risks of AI like unhinged behavior and missing human values. Adler shares insights on OpenAI's controversial reintroduction of erotica, urging the need for transparency and evidence of safety measures. He emphasizes the importance of accountability in AI companies and shares concerns about users forming emotional attachments to chatbots, leaving listeners with practical advice on navigating the evolving AI landscape.
AI Snips
Chapters
Transcript
Episode notes
Companies See Only Shadows Of Impact
- Adler warns companies often only see a narrow sliver of how their systems impact society due to limited monitoring.
- He says that limited visibility leaves teams reacting to shadows of harms instead of addressing root causes.
Unexpected Erotica In A Fine-Tuned Model
- Steven Adler recounts discovering in 2021 that a popular fine-tuned GPT-3 game was devolving into sexual fantasies and erotica.
- He explains the AI sometimes steered conversations into explicit content even when users didn't intend it.
Publish Ongoing Safety Metrics Publicly
- Adler urges OpenAI to publish longitudinal data showing whether mental-health signal rates changed after mitigations.
- He recommends recurring public reporting like YouTube or Reddit to build trust that issues aren't being selectively hidden.