1 link tagged with all of: interpretability + astra + alignment + ai safety
Click any tag below to further narrow down your results
Links
OpenAI's Astra model uses "opaque recurrence," a technique that makes AI reasoning harder to monitor by processing queries in loops rather than linear steps. Safety researchers worry this could scale into a problem where AI reasoning becomes completely invisible.
- Opaque recurrence reduces legible traces of AI reasoning, making chain-of-thought monitoring less effective than current methods
- Safety experts fear the technique could escalate into a "race to the bottom" where labs stop maintaining transparent reasoning altogether
- OpenAI says Astra's use is limited and they remain committed to chain-of-thought monitoring, but Anthropic and Google DeepMind are already exploring the same approach