On the topic of sycophancy: lately I have been basically exclusively using the “thinking” ChatGPT models, o3 and o4-mini-high, and, although the resulting outputs are of higher quality, I noticed that sometimes I feel a sting of missing the “Great idea!” / “What a great question” / … responses of 4o. The “coldness” of the response is especially visible when reading the CoT summaries, including sentences like “the user is aiming to… ” / “it sounds like the user wants...”.
On the topic of sycophancy: lately I have been basically exclusively using the “thinking” ChatGPT models, o3 and o4-mini-high, and, although the resulting outputs are of higher quality, I noticed that sometimes I feel a sting of missing the “Great idea!” / “What a great question” / … responses of 4o. The “coldness” of the response is especially visible when reading the CoT summaries, including sentences like “the user is aiming to… ” / “it sounds like the user wants...”.