Hacker News Reader

Very interesting!

Quick (but difficult) question: do you foresee there arising a reasonably reliable way to filter out the coming wave of ai spam? I’m told that half of Twitter is bots talking to each other at this point, and I’m sure this is coming to other media as well. Eg, massive, massive waves of content marketing, sock puppets, etc.

Is there reason to be optimistic that you or other actors will be able to sift through it?

View on HN

Yes, significant part of our effort is to build technology that detects LLM spam. We have a working model that detects LLM generated text with 90% accuracy currently. The plan is to integrate in search results and make available as an API.

And just out of interest: what is the rate for false positives? An accuracy of 90% sounds very decent, as long as it doesn’t also filter out 10% of legitimate content.

10% of infinity is infinity.

That's as good as 0% accuracy.

And AI text will keep changing.

Reply on news.ycombinator.com