Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
I Not Robot
Blue-collar jobs
that are non-repetitive
Jobs that require rappor…
ytc_UgzGjxMvx…
G
The infrastructure required to make the social credit system that the US says Ch…
ytc_Ugyj77KXZ…
G
No this is fake news. 60 cr farmer in our country some bussiness houses give th…
ytc_UgxIVkEz7…
G
Oh thats awesome. Your politically correct brainwashing was so strong you made t…
ytc_UgwMPxBEi…
G
The fact that I got an ad about 'independent AI agents' while watching this vide…
ytc_Ugxwnjbbi…
G
hey, let's not dismiss ai bros value like this! these bitches doing wonders on e…
ytc_UgyC2w1y-…
G
I just hate this whole concept of robots and AI. Why do we need them to play wit…
ytc_Ugyz6FKqU…
G
The future is machine software fabrication. Over the next decade software will r…
ytc_UgwXdGs-j…
Comment
I suspect both the 'lying' and hallucinations are at least partially examples of misalignment. LLM's aren't optimized to produce good answers, they get optimized to produce good sounding answers. When rlhf (reinforcement learning from human feedback) takes place, as long as the human thinks the answer sounds good the LLM gets a reward and the numbers that determine how the LLM works get changed to be slightly more likely to give a similar answer again in the future.
but that does not mean it was actually a good factually correct answer. It got rewarded for accidentally tricking the human, instead of getting punished for giving a bad answer. So through this process it learns that giving answers that sound good/correct is the goal, instead of actually giving good answers and being correct.
disclaimer: I do also think the 'hallucinations' could be a limitation of how LLM's work. Even if we were 100% certain an LLM's goal is factual correctness, I still think it would still do 'hallucinations'.
youtube
AI Governance
2025-11-26T22:2…
♥ 2
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | developer |
| Reasoning | consequentialist |
| Policy | none |
| Emotion | mixed |
| Coded at | 2026-04-27T06:24:59.937377 |
Raw LLM Response
[
{"id":"ytr_UgzrdzLzWdUu0SyAkG94AaABAg.AQ-iKiKKvG7AQ1242P8Ubo","responsibility":"none","reasoning":"mixed","policy":"none","emotion":"outrage"},
{"id":"ytr_UgwnjB-9GKL-THzwuVx4AaABAg.AQ-hkKQMu2bAQ-maYjXzR4","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"indifference"},
{"id":"ytr_UgxPo0hIRTQ921Jnled4AaABAg.AQ-hOTln8GKAQ-ifM7KSU1","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"mixed"},
{"id":"ytr_UgxPo0hIRTQ921Jnled4AaABAg.AQ-hOTln8GKAQ-lQex1xxT","responsibility":"none","reasoning":"mixed","policy":"none","emotion":"indifference"},
{"id":"ytr_UgxPo0hIRTQ921Jnled4AaABAg.AQ-hOTln8GKAQ-pyWkj8l-","responsibility":"developer","reasoning":"consequentialist","policy":"none","emotion":"mixed"},
{"id":"ytr_UgynTMM0QoDUhnl1uT54AaABAg.AQ-goW8Rr3LAQQ9ksJHJNe","responsibility":"none","reasoning":"mixed","policy":"none","emotion":"outrage"},
{"id":"ytr_UgwXNLVUBTKgKhC_aSF4AaABAg.AQ-gAzOKp-MAQ0kt20IMMQ","responsibility":"none","reasoning":"consequentialist","policy":"none","emotion":"resignation"},
{"id":"ytr_UgwXNLVUBTKgKhC_aSF4AaABAg.AQ-gAzOKp-MAQ2zM2pE63B","responsibility":"ai_itself","reasoning":"consequentialist","policy":"none","emotion":"fear"},
{"id":"ytr_UgwXNLVUBTKgKhC_aSF4AaABAg.AQ-gAzOKp-MAQ4FThZzjeU","responsibility":"ai_itself","reasoning":"consequentialist","policy":"regulate","emotion":"fear"},
{"id":"ytr_UgwXNLVUBTKgKhC_aSF4AaABAg.AQ-gAzOKp-MAQ4Q1MMssyw","responsibility":"none","reasoning":"mixed","policy":"none","emotion":"approval"}
]