This was in their latest announcement: https://cdn.openai.com/spec/model-spec-2024-05-08.html https://cdn.openai.com/spec/model-spec-2024-05-08.html It's quite

This was in their latest announcement: https://cdn.openai.com/spec/model-spec-2024-05-08.html https://cdn.openai.com/spec/model-spec-2024-05-08.html It's quite a breakthrough in how models behave. Tighter security controls, and yet it adds a lot more situations where the AI shouldn't stonewall. It's also reducing refusals that sound judgemental and such.