OpenAI's new reasoning technique alarms safety experts
Opaque recurrence makes a model's reasoning materially harder to monitor.
Opaque recurrence makes a model’s reasoning materially harder to monitor.
Risk 4 · Mitigation 0.
Opaque recurrence makes a model's reasoning materially harder to monitor.
Opaque recurrence makes a model’s reasoning materially harder to monitor.
Risk 4 · Mitigation 0.