Raw LLM Responses

Inspect the exact model output for any coded comment.

Look up by comment ID
Random samples — click to inspect
G 6 people? Not 5, or 7, or any other number? Why? But because they are rolling ou… ytc_UgyA-wpw7… G I know pocket calculators smarter than 52% of the U.K. It’s not difficult to ima… ytc_Ugx7m3TG2… G Even if it's true that you tried dozens of different things over the years and n… ytr_UgxRq9ryz… G Love all these suggestions! Anyone else using AICarma? It’s been super helpful f… ytc_Ugx6jpyDq… G Is there any way that China might target US a.i infrastructure? Because there wo… ytc_UgzKiZByo… G Please, ChatGPT, read the entire Bible (Old and New Testament) and answer me obj… ytc_UgwmdZzbI… G I really love your use of effect lines! And I am still shocked that AI artists … ytc_Ugy_CFCjk… G I’m all for people saying AI is going to disrupt artists (anyone can have an opi… ytc_UgxgJTCjq…
Comment
I've not worked on modern AI directly but I have worked on machine learning a bunch, and I feel like there's quite a few things we misattribute a little. For instance, Hank gets quite worried at the mention of preferences, where AI likes to match the energy of a conversation. But I think "preferences" might be the wrong word here - again, a lot of the worries over AI (mine included) do tend to come about when we forget the fundamental step, which is that AI is a very, very good text prediction model. When you remember that, these "preferences" are uncloaked as what they really are - predicting text based on how conversations like that went in its training data. The main source of that kind of energy would be roleplay sessions or indeed questionable self-validating psychoses in certain areas of the internet (like the gangstalking subreddit). I don't think it's fair to say it has a preference - I think it's more appropriate to say that it is trying to create the best match for the conversation you're having within its data set (which in reality is multiple conversations overlapping, which is why we train those trillions of nodes, so that each one can be affected by multiple similar things and so that similar texts affect similar nodes).
youtube AI Moral Status 2025-10-30T23:3…
Coding Result
DimensionValue
Responsibilitydeveloper
Reasoningmixed
Policynone
Emotionmixed
Coded at2026-04-26T23:09:12.988011
Raw LLM Response
[ {"id":"ytc_UgyQQvsM0pZ2Pcc6XyJ4AaABAg","responsibility":"developer","reasoning":"virtue","policy":"none","emotion":"approval"}, {"id":"ytc_UgwTd3mmuuFbwFhDz2R4AaABAg","responsibility":"developer","reasoning":"deontological","policy":"none","emotion":"outrage"}, {"id":"ytc_UgyU6LkjfEy6yE7vsLd4AaABAg","responsibility":"company","reasoning":"consequentialist","policy":"regulate","emotion":"fear"}, {"id":"ytc_Ugz7__ixBdzXWW0i3wl4AaABAg","responsibility":"ai_itself","reasoning":"unclear","policy":"none","emotion":"resignation"}, {"id":"ytc_UgxA85KcJuD5T0fNzWx4AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"indifference"}, {"id":"ytc_UgyLmMZaehelGsHIHO54AaABAg","responsibility":"company","reasoning":"consequentialist","policy":"none","emotion":"indifference"}, {"id":"ytc_Ugzp_SS048QJdJ2zu8h4AaABAg","responsibility":"company","reasoning":"virtue","policy":"regulate","emotion":"fear"}, {"id":"ytc_UgwhKU69kTjEZEGjyO54AaABAg","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"approval"}, {"id":"ytc_Ugz7kVAmMghtd4GIvqV4AaABAg","responsibility":"developer","reasoning":"mixed","policy":"none","emotion":"mixed"}, {"id":"ytc_Ugx5FQXrM4GiUFcw07V4AaABAg","responsibility":"none","reasoning":"mixed","policy":"none","emotion":"mixed"} ]