cdevroe, 15 days ago

@simon I had the new model and features for a few hours, then it reverted. Must be rolling it out in some odd way.

simon, 15 days ago

@cdevroe by "reverted" do you mean the voice mode stopped working, or that you had the spooky Her voice mode and then it went back to the older, less capable voice mode?

cdevroe, 15 days ago

@simon The icon on the lower right was a new voice mode, and the answers were clearly from the new model. I'm unsure if the voice was new, I hadn't used voice at all with ChatGPT prior. Now the icon is switched back and only text really works (though it will transcribe my voice, but so does iOS).

simon, 15 days ago

@cdevroe that wasn't the new GPT-4o multi-modal voice feature, it was the previous feature where the model (in this case GPT-4o) generates text which is then piped to their tts-1 text-to-speech model

This has confused basically everyone!

cdevroe, 14 days ago

@simon something else I noticed during their demo, they always asked ChatGPT how they were. Even when they had already done so. Perhaps that was a queue to use the new model?

simon, 14 days ago

@cdevroe I think that was more them trying to show off the creepy thing where it pretends to be a person, they got VERY excited about that movie Her

kellogh, 16 days ago

@simon it’s gpt4o for the text though, right? just not audio…

simon, 15 days ago

@kellogh yes, image input and text input and text output are all the new GPT-4o model, which I think is a meaningful improvement on its own