Think about how often you see a question raised and then responded with “I don’t know” in the internet data.
Wikipedia, textbooks, book books, forums, wiki how…
The vast majority of time, the format of what you see is:
<short question>?
<long winded answer or how to guide>
And thats what the llms are all trained on, that’d the most common pattern, so llms replicate that pattern.
The majority if their training data is off wikis abd textbooks, which pretty much never have “I dunno” anywhere. If an answer isn’t known, instead, the page simply doesn’t exist in the first place.
Think about how often you see a question raised and then responded with “I don’t know” in the internet data.
Wikipedia, textbooks, book books, forums, wiki how…
The vast majority of time, the format of what you see is:
<short question>? <long winded answer or how to guide>And thats what the llms are all trained on, that’d the most common pattern, so llms replicate that pattern.
The majority if their training data is off wikis abd textbooks, which pretty much never have “I dunno” anywhere. If an answer isn’t known, instead, the page simply doesn’t exist in the first place.