Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Humans can even agree on the future we want so how in the heck will AI be able t…
ytc_UgwVJgP0k…
G
I don't get it, is the video an hypothetical of technology in the future or aski…
ytc_Ugy2qq0mb…
G
Australia led the way on this type of school. I have family members going. The k…
ytc_UgycVRPjl…
G
Ai cannot replace doctors or lawyers.
Both of these jobs require nuance that an …
ytr_UgwiUszmw…
G
Amazing coverage, John. Powerful essay at the end - I would suggest clipping jus…
ytc_Ugwv85jUq…
G
Guys do you remember when speech to chat first rolled out? Your grandpa talk-scr…
ytc_Ugzzp4FQR…
G
I liked this video very much. It was a device that the kids used to do when the …
ytc_Ugw6dBTRA…
G
It's way more than 40% if hes only counting these "basic to semi complex" jobs. …
ytc_UgxOnwtr1…
Comment
Ok, I went back and read the Nightshade paper (Shan et al), and the authors know what they're doing, it's actually pretty cool - basically, modern AIs work by converting collections of pixels into a simplified 'latent space' where the concepts behind the art are grouped together as detected by a different image-to-text AI model, and Nightshade works by finding a way to trick that model into putting the image into completely the wrong place in the latent space, so conceptually it's doing something similar to what I suggested mislabeling the art but much more efficiently and effectively - against this particular AI. In the long term, though, I still think that if you trained a new image-to-text model from scratch using this data, which to be fair none of these companies are doing because it'd be way too expensive, then these poisoned images would be valuable in the way I assumed they would be before.
youtube
Viral AI Reaction
2024-10-20T21:2…
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | none |
| Reasoning | unclear |
| Policy | none |
| Emotion | approval |
| Coded at | 2026-04-27T06:24:53.388235 |
Raw LLM Response
[
{"id":"ytr_UgwjgT2xgHcxGJIkDRV4AaABAg.A9pVUyKSjzbAA4P1oZS8cl","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"},
{"id":"ytr_UgwxH1gLqg1lpbIx9yJ4AaABAg.A9pVHjb2kU9A9q-FeUv4ds","responsibility":"ai_itself","reasoning":"virtue","policy":"none","emotion":"outrage"},
{"id":"ytr_UgwxH1gLqg1lpbIx9yJ4AaABAg.A9pVHjb2kU9A9qfSoLdpC_","responsibility":"developer","reasoning":"virtue","policy":"none","emotion":"indifference"},
{"id":"ytr_UgwxH1gLqg1lpbIx9yJ4AaABAg.A9pVHjb2kU9A9rQsNgzoBq","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"},
{"id":"ytr_UgzIlMgjZPi0v6q9HTB4AaABAg.A9pV5dxGYTMA9pbEU1SPxq","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"approval"},
{"id":"ytr_Ugx3xiS8IcOSsNDplwl4AaABAg.A9pUyNe5IV5A9p_KVWfh1x","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"indifference"},
{"id":"ytr_Ugx3xiS8IcOSsNDplwl4AaABAg.A9pUyNe5IV5A9pbLeixia2","responsibility":"company","reasoning":"consequentialist","policy":"regulate","emotion":"outrage"},
{"id":"ytr_Ugx3xiS8IcOSsNDplwl4AaABAg.A9pUyNe5IV5A9pdhC79g-x","responsibility":"developer","reasoning":"deontological","policy":"none","emotion":"outrage"},
{"id":"ytr_Ugz6QScWr6E3XxX7F5J4AaABAg.A9pUwgo0NfFAG_u0vQIG9r","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"outrage"},
{"id":"ytr_Ugw1VVFhwGjEeEskUGV4AaABAg.A9pUc8wywZCA9pjcgmb8PG","responsibility":"government","reasoning":"consequentialist","policy":"ban","emotion":"outrage"}
]