There's probably a sweet spot here. On the flip side, ChatGPT currently doesn't indicate whether a given image generation request was serviced by multimodal GPT-4o [1] or Dall-E.
Personally, I do like the "use web search" and "extended thinking" buttons, but ultimately, the models should probably be able to figure out whether doing so would be useful themselves too.
Personally, I do like the "use web search" and "extended thinking" buttons, but ultimately, the models should probably be able to figure out whether doing so would be useful themselves too.
[1] https://news.ycombinator.com/item?id=43474112