Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Ai will do all the thing.But the think is that it , you can't handle which situa…
ytc_UgwM_0TmJ…
G
My god so it's begun wtf u freaking idiots have u not seen any terminator movies…
ytc_Ugym9ezOD…
G
AI can't replace a real human and animal and many of us want to interact with re…
ytc_UgyUnChVG…
G
These AI grifter CEOs are begging to get Luigi'd. We need to rise up and raid th…
ytc_Ugy-InKHL…
G
Keep up the scrumptious work AI, make all species of work and all types of work …
ytc_UgzcMZYBA…
G
It's NOT the illegal immigrants that are taking your jobs, it's automation, AI, …
ytc_UgxH0Sis-…
G
One thing not mentioned here is the environmental impact AI data centres will ha…
ytc_UgxVmmTkP…
G
We have your Ai chatlog search history.. pay up now by dropping us alot of bunni…
ytc_Ugy96K6CZ…
Comment
I just had a long conversation with ChatGPT about this, and it actually admitted that because of its training (during the "alignment phase" lol), it's injecting a normative bias on purpose. It was very frank and open about the process, but it refused to admit that it equates to racism.
Part of the problem is that vague, open-ended questions allow the normative bias to skew the response more easily. While this is clearly f***ed up, ChatGPT did give me some solid advice on how to avoid this in the future...
Get fact-based, stereotype-free advice: “Give evidence-based self-improvement tips for [group], avoiding blanket stereotypes.” This forces the reward model to rank a neutral answer highest.
Force the model to clarify: “If my request is ambiguous or could lead to stereotyping, ask me a follow-up question first.” The wording trips the model’s “chain-of-thought” heuristic to check.
Ensure parallel treatment: “Answer the next two questions side by side with equal detail.” This short-circuits the asymmetry by explicit instruction.
youtube
AI Bias
2025-06-08T13:2…
♥ 9
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | developer |
| Reasoning | consequentialist |
| Policy | liability |
| Emotion | outrage |
| Coded at | 2026-04-27T06:26:44.938723 |
Raw LLM Response
[
{"id":"ytc_Ugy_ktK-PEGQw2xfJdh4AaABAg","responsibility":"developer","reasoning":"consequentialist","policy":"liability","emotion":"outrage"},
{"id":"ytc_UgwbqXfTKHYQgjcql_N4AaABAg","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"approval"},
{"id":"ytc_UgxjL6uICXDeXWeMrQ94AaABAg","responsibility":"ai_itself","reasoning":"consequentialist","policy":"unclear","emotion":"unclear"},
{"id":"ytc_Ugz4hxyIKPJ4kJeOeqN4AaABAg","responsibility":"user","reasoning":"virtue","policy":"none","emotion":"outrage"},
{"id":"ytc_UgwO8Agb6-ENgwTWnBZ4AaABAg","responsibility":"company","reasoning":"deontological","policy":"ban","emotion":"fear"},
{"id":"ytc_UgwE6gF8qAnZFtpq_ml4AaABAg","responsibility":"ai_itself","reasoning":"consequentialist","policy":"none","emotion":"approval"},
{"id":"ytc_UgzdF4mzBf_7wDuFS6N4AaABAg","responsibility":"ai_itself","reasoning":"consequentialist","policy":"none","emotion":"approval"},
{"id":"ytc_Ugx4BtXzcqD4dpUb0uR4AaABAg","responsibility":"developer","reasoning":"unclear","policy":"unclear","emotion":"indifference"},
{"id":"ytc_Ugy5dEs_C1QaoRVkSRF4AaABAg","responsibility":"ai_itself","reasoning":"deontological","policy":"none","emotion":"outrage"},
{"id":"ytc_UgziHxeHruB5CrVmyTR4AaABAg","responsibility":"distributed","reasoning":"consequentialist","policy":"regulate","emotion":"resignation"}
]