Raw LLM — Corpus Dashboard

Look up by comment ID

Random samples — click to inspect

G I imagine this going in the same way as the case as the chimp taking a picture w… ytc_Ugz5GTC54… G No I need to speak with real people it’s so hard to talk to someone real at xfin… ytc_Ugwlbjb9r… G AI programed to be more human. This is normal human behavior. Proceed to cut the… ytc_UgwzslA-l… G You can do a lot of different things with the same search index. The algos for r… rdc_n3y4j5l G They’ll always need humans if we just refuse to use AI to do everything for us. … ytr_UgxGv6RAf… G Dear Corporations:) AI is A LANGUAGE MODEL!….It is ‘intelligent’ in the same way… ytc_UgwI2Qgg6… G These are simply parsed scripts with a random element thrown in. A random elemen… ytc_UgwmNSHMM… G ⁠@aleksipekkala By every technical definition it is art. The definition of art i… ytr_UgwXWSs3c…

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"}
]

Raw LLM Responses