Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
I cant see a way where Putin would get easily part of Ukraine. This would surely…
rdc_cfkxff0
G
I agree mostly with all the points you made until 13:25 .
Automation isn't a bla…
ytc_UgyfB3T7r…
G
Ai is still horrible whether you're lying about it or not. It steals from other …
ytr_UgxOWTVEm…
G
कातिल पत्नी से बचाव का replacement 😂 न प्यार, न धोख़ा,,, अब शादी करने और gf की ज…
ytc_Ugx8usdeG…
G
The only thing that will stop this is if people stop being greedy and worshippin…
ytc_UgytmusEa…
G
These companies have gone to far. I hate AI. I really hate calling and getting a…
ytc_Ugyu57LVq…
G
Looking at anything AI generated feels like looking at either stolen work, or th…
ytc_UgxAEowrd…
G
New regulation is needed for AI. Any AI generated images or videos should be exp…
ytc_Ugwdq_0Bs…
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"}
]