there seems to be a "new" GPT-4o in ChatGPT -- Eg. the model feels completely different in how it reasons and responds -- sometimes it feels worse, other times it just feels different. Any thoughts?
Yes, OpenAI recently introduced GPT-4o, and many users have noticed differences in its reasoning, response style, and overall feel compared to previous versions. Some say it feels faster and more conversational, while others feel it occasionally simplifies things too much or changes its approach unexpectedly. These differences likely stem from optimizations in efficiency, response balance, and alignment with real-time interaction goals.
The chatgpt-4o-latest endpoint has branched away from the 4o-08 (and/or 4o-11) models quite a while ago, so it's not unlikely they've updated it again.
Sonnet has also had silent changes, likely at the system prompt level - not the public one, the non-public "alignment" one that is even applied to the API - that have made it lazier in terms of maximum output length (despite max_tokens values).
Yes, yesterday I accidentally used the 4o instead of 4.5 model and noticed it was behaving differently. Showed me a table with pros and cons, suggested me to generate a document that I could print for better convenience etc. Seems it's better than before but haven't used it that much to be sure.
Sonnet has also had silent changes, likely at the system prompt level - not the public one, the non-public "alignment" one that is even applied to the API - that have made it lazier in terms of maximum output length (despite max_tokens values).