back
1 comments
Importantly, we cannot trust in "competition" between poisoners to save us, because every additional spammer simply buries the real truth (or at least, good-faith data) further down into obscurity.

Worse, you don't have to use it to suffer from it. This form of "AI search" sets up every other site (e.g. HN) as a battleground between bots/scammers/astroturfers in order to indirectly manipulate results.

> language models use lexical similarity to a query as a stand-in for accuracy of information

This framing can be extended to (or perhaps from) the issue of "prompt injection". It's all injection all the time. Your words aren't convincing a mind into accepting a fact through logic (or even obscure human intuition) as much as shunting the text-similarity train onto a new path.