Re:ChatGPT has trouble giving an answer before explaining it...
by refulgentis on Tuesday March 07, 2023 @07:39PM
(#35063600)
yes, spent a fun couple weeks layering this on top of GPT-3.5. Hopefully I'll write it up soon.
I'm a bit embarrassed to: in the process, I picked up a habit of reading research on prompting. "Real" research finetunes internal models to play a particular role, rather than orchestrating several "conversations" and hoping your prompt will get you the right output format 100% of the time, etc.
I'm a bit embarrassed to: in the process, I picked up a habit of reading research on prompting. "Real" research finetunes internal models to play a particular role, rather than orchestrating several "conversations" and hoping your prompt will get you the right output format 100% of the time, etc.