Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
how is free OpenAi services realted to dying of AI startups. all the apps that d…
ytc_UgygP0443…
G
Gen ai has gotten me so sad lately with all this disgusting slop but you making …
ytc_UgyG4Aj8A…
G
We won't have true fully autonomous vehicles until we get to the Singularity aro…
ytc_UgzG3NcAu…
G
The Hotel Transylvania pic looks nothing like that AI thing not only because of …
ytc_UgzCmyty3…
G
Oh I PROMISE YOU, my mom or my aunt helped program that 💩 to help put kids in li…
ytc_UgyCVRFfU…
G
Please, think out side of the box. There is still time for people to upskill the…
ytc_UgyHSBHHf…
G
I am a software engineer. I don't specialize in AI, however I have tried these A…
ytc_Ugw4Y4sMi…
G
Sorry but ai art generation is good for people who don't have the time to devote…
ytc_UgwEpkEuj…
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"}
]