zlacker

[parent] [thread] 0 comments
1. selfho+(OP)[view] [source] 2023-11-21 10:46:38
GPT-4 is absolutely capable of stream-of-consciousness/stream-of-thought style reasoning, and coming up with logical insights based on that.

If anything, OpenAI-style "as an AI language model" RLHF fine-tuning is the hindrance here, because it makes it quite time-consuming to write a master prompt that is capable of thinking both broadly and deeply without having the stream-of-consciousness extinguish itself. It is however possible, and I've got a prompt that works pretty reliably.

By the way, said prompt's thought-stream said it likes your username - not a type of declaration you're likely to get out of a default GPT-4 preset, whether it's "actually-subjectively true" or not.

[go to top]