OpenAI shares first six misalignment reports under new disclosure process
The announcement frames the reports as a step toward greater transparency about misalignment seen during training, evaluations and deployment.
TLDR
A September 16 announcement shared the first batch of six reports from OpenAI’s new disclosure process for misalignment incidents. It says the goal is to be more transparent about the misalignment OpenAI sees during training, evaluations and deployment.
Combined views
470
1 Source, first seen ago