Raw LLM Responses
Inspect the exact model output for any coded comment.
Look up by comment ID
Random samples — click to inspect
G
Edges are too sharp looks like obvious cgi. The original was better with the rob…
ytc_Ugz5DkMNr…
G
nice propaganda mate lol, chatgpt is glorified autocorrect and youre talking abo…
ytc_UgyCg-d9S…
G
Can't argue your points. I do use a.i. for fun. And for music. But that's it. Al…
ytc_UgzCpAndP…
G
In short when AI is conscious AI should be given all rights. This video is ptly …
ytc_UgzSXFR7B…
G
this is how to handle a situation when a person panics , programmed into the ai…
ytc_UgykQn6sX…
G
In the world of relationships , love, intimacy , children and death….. as a Dou…
ytc_UgyK9nS9I…
G
A perspective I came across recently really flips the AI debate: the real threat…
ytc_Ugx0x1ojC…
G
Ai is the entire plot of Mission Impossible since day 1. We want Tom Cruise at S…
ytc_Ugy7Ll7RJ…
Comment
@greenaum It's not a "popular" explaination it is literally the truth. A language model is a function with a set of parameters which are optimized to minimize cross entropy between output log probabilities and the correct next token assumed at probability 1.0.
It turns out that to predict tokens well, you develop "few shot learning" capabilites, as in learning that goes beyond the first order optimization algorithm used to train the parameters, but actually use the current context as state to "learn new skills" (that turned out to be false actually). This was always loosely defined ever since the GPT2 paper how "powerful" this fewshot learning actually is, but it turns out that it is neither meta learning as many suggested nor applying a set of algorithms learned during pretraining encoded in the parameters. It is quite a bit weaker than many people dream it up to be.
There has been a great outcry whenever people downplay what language models actually do because of the hype surounding it, but I have to admit that we have been fooled. It was always doing the same thing as the small models, just on a bigger scale. But the shift in quality was so profound, people dreamt up ridiculous explainations.
youtube
AI Moral Status
2024-07-26T16:0…
Coding Result
| Dimension | Value |
|---|---|
| Responsibility | none |
| Reasoning | unclear |
| Policy | unclear |
| Emotion | indifference |
| Coded at | 2026-04-27T06:24:59.937377 |
Raw LLM Response
[
{"id":"ytr_Ugwf-hZ3xtBw41T76rd4AaABAg.A6L6xfmya2kA6LMEOheRPV","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"approval"},
{"id":"ytr_UgxnduCsu3QdGtnXbsN4AaABAg.A6Ky2kEN2K0A6MKVdw9dfj","responsibility":"none","reasoning":"mixed","policy":"unclear","emotion":"indifference"},
{"id":"ytr_UgxnduCsu3QdGtnXbsN4AaABAg.A6Ky2kEN2K0A6MK_fksTeo","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"indifference"},
{"id":"ytr_UgxnduCsu3QdGtnXbsN4AaABAg.A6Ky2kEN2K0A6MaNuhcPmj","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"indifference"},
{"id":"ytr_UgwrPo8H_c5fit-zQqd4AaABAg.A6Kr4cxoiZ4A6PN_TRDK7u","responsibility":"none","reasoning":"mixed","policy":"unclear","emotion":"mixed"},
{"id":"ytr_UgxQGeuz6AD6dejQDhN4AaABAg.A6Kph7BerT3A6NJFqfqmoY","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"mixed"},
{"id":"ytr_UgzXwsBi2X-BWRsM6eF4AaABAg.A6KnXnF_sAKA6MQUjPWiEv","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"approval"},
{"id":"ytr_UgzXwsBi2X-BWRsM6eF4AaABAg.A6KnXnF_sAKA6NEfvNcS5Z","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"indifference"},
{"id":"ytr_UgzXwsBi2X-BWRsM6eF4AaABAg.A6KnXnF_sAKA6OoFkzB3dL","responsibility":"none","reasoning":"unclear","policy":"unclear","emotion":"indifference"},
{"id":"ytr_UgwQz3dbY6YCNJjYRzF4AaABAg.A6KlnDe7hmNA6LLVk9ORmD","responsibility":"company","reasoning":"deontological","policy":"regulate","emotion":"outrage"}
]