alt Hacker NewsBut this only seems wrong to you because you're familiar with the prior context. When there's only a single paragraph to work from, and it's a drily humorous text, why not employ comic literalism and lean into the perplexity with which his neighbors viewed him?
Replies
mvdtnz • yesterday at 8:52 PM
The LLM has access to the full text and every translation of it ever produced, both in its corpus and on the searchable web.
lern_too_spel • yesterday at 9:40 PM
The fact that a ring appears at the end proves that the model has context beyond what was in the prompt. The model clearly knows the story that the prompt was extracted from.
emp17344 • yesterday at 9:53 PM
This is a rationalization. The simpler explanation is that the model failed to properly depict the passage. As the AI booster crowd would say, cope.
Isn't this almost certainly in the training set?