Post by Aaron Mills 🏳️⚧️
The only way to trust an LLM not to do something is to not give it the ability to do that thing in the first place. Otherwise it may absolutely do that thing. Even if you shove a safety classifier in…
The only way to trust an LLM not to do something is to not give it the ability to do that thing in the first place. Otherwise it may absolutely do that thing. Even if you shove a safety classifier in…