|
|
|
|
|
|
by Terr_
755 days ago
|
|
|
> The LLM is like another user. I like to think of LLMs as client-side code, at least in terms of their risk-profile. No data you put into them (whether training or prompt) is reliably hidden from a persistent user, and they can also force it to output what they want. |
|
|