On: ChatGPT-4 significantly increased performance of b...

[bryancoxwell]

I’ve also found the act of describing my problem to GPT4 is sometimes just a helpful as the answer itself. It’s almost like enhanced rubber duck debugging.

[airstrike → bryancoxwell]

So true. I've written entire prompts with several lines worth of explanation, only to realize what my issue was and never hit the "send" button. Guess I should do that more often in life, in general

[dkjaudyeqooe → bryancoxwell]

We need an inverse GPT4-style LLM that doesn't provide answers but instead asks relevant questions.

[awestroke → dkjaudyeqooe]

GPT4 can do that too. Just show it something (code or text) and ask it to ask coaching questions about it.

[seanhunter → dkjaudyeqooe]

I have tried adding prompts like this and it works really well. "Rather than giving me the answer, guide me using questions in the Socratic method".