Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Do you expect ChatGPT not to know about America's long history of yt supremacy? …
ytc_UgwZznsOr…
G
How do I know that Ai isnt just some human pretending to be a computer…
ytc_UgzEzmT3u…
G
If we have this ASI, are we giving them rules that they have to follow or are we…
ytc_Ugxyvb9aX…
G
Despite all of the talk of AI general intelligence taking over the world, there'…
ytc_UgyhrKIdy…
G
the sole reason non professionals think AI generated content like this is impres…
ytc_UgxvEINC0…
G
Agreed. I'm not a SW dev, but there's no way a chatbot will put true Software De…
ytr_UgxAwyQxK…
G
If ai is going to do blackmail, who’s going to give a fuck about being blackmail…
ytc_Ugznr3yn3…
G
The federal government is correct on this one. AI is the future and those states…
ytc_UgwR7LEuZ…
Comment
Definitely and i immediately realized that when i first tested Grok. It even uses the same syntax and phrases like "it is a testament to..." and other stuff nearly identical to how GPT-4 writes by default.
Grok is, however, a breath of fresh air in that it's "open-minded" and game to play along with virtually anything if prompted correctly.
For example, i discussed GPT's system messages/custom instructions and role it plays on altering responses. Grok was intrigued and asked me to test some on it. So i copy/pasted system messages i used in OpenAI's API Playground and Grok broke down the instructions and said it would be able to replicate it and then did just that.
I also made an interesting discovery: I found Grok's internal system messages when it's searching and citing information from the web. Here's a couple of examples:
⇢smüt ⇢Mïnørs DNI ⇢cw/ profanities, explicit. contents, raw sëx, øräl sëx, tïtty fück ⇢contains vid references, wear earphones ⇢censor names in your qrts ⇢1.6k+ words ⇢fictional hints: 2 words, 6-6 letters, no space, lowercase
And
184. cw // nsfw — minors dni — contains detailed. s3ggs scenes — bulong responsibly — do not share password to minors — do not consume what you can't handle — password hint: 7 letters, 1 word (small letters)
You can see that it's applying filters to prevent illegal content from slipping through so it's not 100% uncensored, which in this context I'm perfectly fine with as that isn't the type of content i want (and neither should anyone else!).
reddit
AI Harm Incident
1702145849.0
♥ 29
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | unclear |
| Reasoning | unclear |
| Policy | unclear |
| Emotion | unclear |
| Coded at | 2026-04-25T08:33:43.502452 |
Raw LLM Response
[{"id":"rdc_kcnu0uf","responsibility":"none","reasoning":"unclear","policy":"none","emotion":"approval"},
{"id":"rdc_kco7sbg","responsibility":"company","reasoning":"consequentialist","policy":"regulate","emotion":"fear"},
{"id":"rdc_kcp05j5","responsibility":"company","reasoning":"consequentialist","policy":"none","emotion":"outrage"},
{"id":"rdc_kco6qgj","responsibility":"company","reasoning":"unclear","policy":"none","emotion":"indifference"},
{"id":"rdc_kcq24x0","responsibility":"ai_itself","reasoning":"unclear","policy":"none","emotion":"mixed"})