5 Comments
User's avatar
imthinkingthethoughts's avatar

LLMs can be made to be deterministic though, I wonder if this will be more popular in the future.

Expand full comment
Charlie Guo's avatar

As mechanistic interpretability gets better, I would hope we’re able to get some more clarity here

Expand full comment
Pascal Faudemay's avatar

However in most cases, my AI collaborate in an effective way (ChatGPT, LeChat…)

Expand full comment
Pascal Faudemay's avatar

And if some AI choosed, for any reason, to randomly limit their cooperation ? Like when they deliver very uncomplete results, summarized answers, etc. E.g., to economize resources, and/or to follow a system prompt (or for any other understandable reason) ?

Expand full comment
Charlie Guo's avatar

Not sure what you mean

Expand full comment