Does ChatGPT provide better results if you let it think out loud first before giving the final answer?

Prompting
1

I think I read somewhere that the rambling nature of ChatGPT which many people don’t like is actually something that improves the quality of the answers, because you give it a chance to think out loud.

I want to query structured output through the API, but I’m wondering if it’s better to let it answer in free form so that it can think out aloud first, and then ask it to give the final answer in some distinct way such as wrapped into a html tag, so that it can be parsed. The structured output response_format does not have this option though (I guess?), so I wonder if you’re making some sacrifices in response quality there.

Any thoughts?

1 Like
2

Yes if you let the GPT list out its response in steps it is more accurate, it’s a model so the more steps it has the more it needs to list them out. Like as an example when I add up open chats if I just copy paste it never adds them up right. But if I have it break the 87 into groups of ten with one group of 7 then total the groups it is more accurate.:rabbit: