← sources

The New "I'll provide responses in chunks" UX is a downgrade

source: reddit (reddit:ChatGPT)published 2026-10-12 08:19ingested 2026-10-13 02:37

retained text

I find the new "feature" where GPT-6 Thinking starts creating an incomplete answer immediately and then updates it with additional reasoning and information it finds really unreasonably annoying and a downgrade compared to be previous model behavior.

It starts with an incompletely reasoned speculative answer, that then evolves throughout the answer it provides, where the conclusion sometimes does not match the initial information provided.

What's worse, it often introduces bias where the model starts from a wrong conclusion that affects the rest of the answer it builds on top of. But it also introduces immediate bias in the reader who first sees a misleading confident answer, and then has to deal with frequent mini-cognitive dissonances of having to immediately change what they believe based on the updated text as it develops. Even though I recognize it, it still feels annoying to my brain to have to re-convince itself to change its belief about the question all the time.

To make that aspect even less pleasant, it also updates the actual content on the screen as you try to read its interim answer, and you aren't quite sure when it's fully done.

Annoyingly, GPT-6 starts giving me often the wrong answer within seconds, which is completely unnecessary. I would much prefer to wait until it's finished its research and reasoning, and started from a complete correct answer.

I truly wish there was a way to disable this. The old "this is what I've found so far, and this is what I'm doing now" status updates were the far better way to go about it.

This is a bit of a rant, but I wonder how do others feel about it? Because I may be crazy and this actually adds value to everyone else. But even then, I wish there was a way to toggle "start seeding partial response early" vs "provide full, coherent answer once you've finished collecting data and reasoning". 

document fetched 2026-10-13 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1x3w9yl/the_new_ill_provide_responses_in_chunks_ux_is_a