Comment by peab

I think it's more so that they push changes quickly without exhaustively testing. Compare that to Google, who sits on a model for years for fear of hurting their reputation, or OpenAI and Anthropic who extensively red teams models