Raw LLM Responses

Inspect the exact model output for any coded comment.

Look up by comment ID
Random samples — click to inspect
G Did a lot of delivery pre ai and cameras for a couple companies. I think if you … ytc_UgzRiVwyM… G Here is the key difference however. I think AI art is a form of art it should be… ytc_Ugz0outPl… G ai art is so cpnfusing to me, like, writing is an art form, so does that mean ai… ytc_Ugzoi_SJG… G It's a relief to hear the AI Safety Expert talk about carbon and Net Zero wrt to… ytc_Ugy0GErmk… G I'm a musician, making music for movies and TV-documentaries. Of course, since … ytc_Ugwvy1g32… G I love how sometimes the red AI said some misinformation or misunderstanding abo… ytc_Ugy2tJ5az… G Just wait until modern media outlets that do surface-level "research" into their… ytc_UgwAUvzSg… G @kitchenersown No, it's just because using ai art for such things might come wit… ytr_UgzwyiQ6U…
Comment
Eliezer had a great example several years ago (when GPT-2 was new) of where being able to predict text makes you need to be much smarter than whoever produced it, which Iʼll adapt here. I have a WARC file on my computer. This is a concatenated set of HTTP requests and responses, each preceded by a bunch of pseudo-headers. If I gave part of it to an LLM base model and ended with WARC-Block-Digest: sha1:C7TER6D5FOEWSUZSTS7W4M37BZDIZLDJ WARC-Payload-Digest: sha1:X3D542ITV7HNYEEMVAH5LHRDPQYFCWFF Content-Type: application/http;msgtype=response Content-Length: then to accurately do the autocomplete it would need to output the number 784975, then an HTTP response which gave the right hashes, which is not a thing that any human could possibly due unless they happened to already know the answer. Even if you had the full context, which would help a great deal, I donʼt know if you could actually do it without finding and looking at the webserver logs.
youtube AI Moral Status 2025-11-09T06:3…
Coding Result
DimensionValue
Responsibilitynone
Reasoningconsequentialist
Policynone
Emotionindifference
Coded at2026-04-26T23:09:12.988011
Raw LLM Response
[ {"id":"ytc_UgzCfgXOWqj_QckvzY14AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"approval"}, {"id":"ytc_Ugy1PcgxyRpO6yFePBd4AaABAg","responsibility":"government","reasoning":"mixed","policy":"regulate","emotion":"mixed"}, {"id":"ytc_UgysOsgfV69frC13hlN4AaABAg","responsibility":"ai_itself","reasoning":"virtue","policy":"none","emotion":"outrage"}, {"id":"ytc_Ugzwr_KSzvipseA0Au94AaABAg","responsibility":"developer","reasoning":"unclear","policy":"none","emotion":"indifference"}, {"id":"ytc_Ugyl9pZdZa4uSa23sUZ4AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"approval"}, {"id":"ytc_UgwB_5LjgvmB9LLCc3d4AaABAg","responsibility":"developer","reasoning":"deontological","policy":"liability","emotion":"fear"}, {"id":"ytc_Ugy03wl9LdwnUgQDn0l4AaABAg","responsibility":"company","reasoning":"mixed","policy":"regulate","emotion":"outrage"}, {"id":"ytc_UgzURJ6yX_tzv56jRcV4AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"indifference"}, {"id":"ytc_UgyySM7JDt6YFvJjZd54AaABAg","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"approval"}, {"id":"ytc_Ugz8jelfArGzHzPt87F4AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"indifference"} ]