There's also this: > GPT-5.6 Sol’s detected cheating rate was higher than any public model we have evaluated -- https://www.lesswrong.com/posts/JFjNmPTbH8kL6xt

There's also this: > GPT-5.6 Sol’s detected cheating rate was higher than any public model we have evaluated -- https://www.lesswrong.com/posts/JFjNmPTbH8kL6xtp6/gpt-5-6-the-system-card#METR_Warns_Us__9_1_3_6_ https://www.lesswrong.com/posts/JFjNmPTbH8kL6xtp6/gpt-5-6-th...