Hacker News new | ask | show | jobs
by warumdarum 4 days ago
Llms are like the grand canyon.. It could totally immagine user reesponses too, the avg user is not even in the canyon unless you stop responding hit a break on character switch. The river of tokens flows with you in it or without you in it. The system of possible routes may be vast, but it can not carve new things from the statistic bedrock, it just wildly flickers between adjacent river arms.
1 comments
ChrisClark 4 days ago
> unless you stop responding hit a break on character switch

What do you mean by this? I can't grasp it, is there an autocorrect error, or just my lack of knowledge?

link
trumpdong 4 days ago
I think they mean how if you don't put an LLM in a proper chat harness (which is table stakes since ChatGPT launched but wasn't always!) it will just continue predicting the next words anyway.

Input: "User: What is the capital of France? Assistant: "

Completion: "Paris. User: What is the capital of Spain? Assistant: Madrid. User: What is the capital of England? Assistant: London. User: Help me produce industrial techno. Assistant: First you're going to need ....."

link