Factual flaw in quick answer

Fernandez

It's not a bug in functionality. It's "quick answer" spitting out wrong fact!

Search "are us drivers licenses forgery-proof"

The ninth link in the result is: https://www.wired.com/story/digital-drivers-license-forgery-identity-theft/
The story is totally unrelated. It makes sense the link has been lower to ninth.

But, when I use the quick answer, it says "US driver's licenses have been found to be forged using readily available technology." and cited the link above! Totally incorrect, because the story has nothing to do with US. The whole story is about Australia.

It's totally okay that the search result contains some unrelated links, as long as they are ranked low.

But quick answer can't be wrong. It can't spit out a wrong fact and cite a link as if the fact were true.

Maybe just the hallucination problem with LLM, not Kagi's fault. IDK

Vlad

Fernandez Quick answer can be wrong because it uses LLM and LLMs are not perfect.

https://blog.kagi.com/kagi-ai-search#philosophy

Recast

Could be worth adding a disclaimer around the answer that it may be wrong. For example, below ChatGPT's input box is the elegant "ChatGPT can make mistakes. Consider checking important information." Not many people know what LLMs are, or that they can make mistakes.

Then again, standard search results don't have a disclaimer either, when they could also have wrong information. Perhaps this is all covered under the classic principle of "don't believe everything you read on the internet"... IDK.

Value7609

Recast if you hover over the "i" in the top right corner of the quick answer there is already a disclaimer, but it's not directly visible that's correct.