Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
rcxdude
4 months ago
|
parent
|
context
|
favorite
| on:
Weaponizing image scaling against production AI sy...
The fact that instruction tuning works at all is a small miracle, getting a rigorous idea of trusted vs untrusted input is not at all an easy task.
cubefox
4 months ago
[–]
It should work like normal instruction tuning, except the SFT examples contain additional instructions in <|quote|> tokens which are ignored in the sample response. So more complex than ordinary SFT but not that much more.
rcxdude
4 months ago
|
parent
[–]
There are LLM finetunes which do this, it is very far from watertight.
cubefox
4 months ago
|
root
|
parent
[–]
Example?
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
