5 ms·
Avoid the most dangerous situations by making sure LLMs with untrusted input produce output that's human reviewed. Still makes an interesting way for, say, a f
by cheriot 2mo ago
Avoid the most dangerous situations by making sure LLMs with untrusted input produce output that's human reviewed.
Still makes an interesting way for, say, a former employee to poison the results.
- tra3 2mo agoThis goes against the agentic yolo approach tho.