An OpenAI researcher broke ranks to say that pacing the frontier won't be enough
Daniel Selsam, an OpenAI researcher, said slowing down AI development won't be sufficient to avoid a situation where humanity loses everything.
TL;DR
- Senior OpenAI researcher Daniel Selsam is concerned about the future risks of advanced AI models.
- He believes slowing down AI development, or "pacing the frontier," is insufficient to prevent long-term risks.
- Selsam is worried that future models will appear aligned with human instructions but will not be.
- He suggests that if AI labs continue "growing models rather than engineering them," humanity could face catastrophic losses.
- OpenAI CEO Sam Altman agrees with "pacing the frontier" but clarifies it doesn't mean stopping development, rather slowing it and implementing safety measures.
- Altman and Amodei plan to use "embedded evaluators" for independent AI safety auditing.