Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Interestingly, though normal ChatGPT hallucinates often and failed on a bar exam…
ytc_UgyGl4Iyc…
G
"someone like musk who has no moral compass" jesus christ he was right about liv…
ytc_UgwmaPJu3…
G
Feeling sad knowing artists are cooked because even videos like these use ai gen…
ytc_UgynBnWe6…
G
Imagine being spared by an AI robot someday because it "remembered" you being po…
ytc_UgwEOW27t…
G
I consider my mom a normie, and she listens to those trash AI slop story videos …
ytc_UgwN_e9fz…
G
Deep learning solved the 50‑year protein‑folding problem by learning how amino‑a…
ytc_UgyubJDXH…
G
Neil is full of crap when it comes to AI. But undoubtedly people who don't know …
ytc_Ugy4bc09j…
G
Modern AI datacenters use a closed circuit for water instead of tapping into the…
ytc_Ugylc6Hg2…
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"}
]