Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Using an LLM for coding is like getting a plumber to come over to install a sink…
ytc_UgzjSR-vh…
G
I think i actually have a different opinion here.
(im writing on the fact of cop…
ytc_UgxFCKpSt…
G
When actual AI is born or when people realise that current AI isn’t actually AI.…
rdc_n7zuwma
G
The winner of AI….likely USA or China want to use AI to control the planet and e…
ytc_UgwRcR3ny…
G
Are we just gonna ignore the fact he said he wants to take over the main power g…
ytc_UgzPjjcLH…
G
1:28:32 I agree that consciousness is probably not a tangible quality, as stated…
ytc_UgxVQto9i…
G
ATS isn't inherently AI based, though rule sets have been around for a long time…
ytr_UgyUKVhsb…
G
Imagine (I guess it is a given) indivicuals use AI and AI agents and other compl…
ytc_UgzrXbdEa…
Comment
honestly I used chatgpt 3.5 a lot, and I believe he probably could have gotten it to say 'why yes, you should go right ahead and eat some sodium bromide, yum yum yum'. but nobody was seriously telling people to talk to AI models for health anything at 3.5.
GPT 4 you could *generally* get sane advice from ChatGPT on most subjects as long as you didn't try hard to push it off track... and imo GPT 5 is significantly safer than google (which isn't to say that ChatGPT should replace a doctor's advice). But with 3.5, I specifically remember getting ChatGPT to endorse some terrible, terrible things by talking to it for long enough. I never tested something exactly like this, so I can't make promises, and it's entirely possible that the AI didn't do anything too crazy here, but I will say that 3.5 was unsafe and there's good reason for OpenAI's discontinuation of it.
youtube
AI Harm Incident
2026-04-21T18:3…
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | user |
| Reasoning | consequentialist |
| Policy | industry_self |
| Emotion | indifference |
| Coded at | 2026-04-27T06:24:53.388235 |
Raw LLM Response
[
{"id":"ytc_Ugyq7F8uKd4-q6H9KVJ4AaABAg","responsibility":"ai_itself","reasoning":"consequentialist","policy":"unclear","emotion":"mixed"},
{"id":"ytc_Ugw-sACa30q38aUCiER4AaABAg","responsibility":"user","reasoning":"consequentialist","policy":"industry_self","emotion":"indifference"},
{"id":"ytc_UgzQVy8xXvsbGgG35HV4AaABAg","responsibility":"user","reasoning":"unclear","policy":"none","emotion":"mixed"},
{"id":"ytc_UgwHXFYLZSlUeXxCJLd4AaABAg","responsibility":"company","reasoning":"deontological","policy":"ban","emotion":"outrage"},
{"id":"ytc_UgyCux2GKQxk0BvIrGx4AaABAg","responsibility":"user","reasoning":"virtue","policy":"none","emotion":"mixed"},
{"id":"ytc_UgzQEUGuAWwaCn8fOFF4AaABAg","responsibility":"distributed","reasoning":"consequentialist","policy":"liability","emotion":"outrage"},
{"id":"ytc_UgwhvOum004-Hp6wjCF4AaABAg","responsibility":"company","reasoning":"deontological","policy":"regulate","emotion":"outrage"},
{"id":"ytc_Ugy3Gknio5-FAbynV4Z4AaABAg","responsibility":"ai_itself","reasoning":"consequentialist","policy":"unclear","emotion":"mixed"},
{"id":"ytc_UgwKcZPI7CfR7CFmqCJ4AaABAg","responsibility":"company","reasoning":"contractualist","policy":"regulate","emotion":"fear"},
{"id":"ytc_Ugxi85BHGv50ld_SYnV4AaABAg","responsibility":"ai_itself","reasoning":"unclear","policy":"none","emotion":"mixed"}
]