Hacker News Reader

It's is more about Google search than it is about the internet.

There was a period in the past when human spam was a problem that was not trivial to solve.

As always, modern problems require modern solutions.

The most effective spam filtering is done not by content but by various white and black lists of providers. Essentially it is a trust score which is a very old solution to a lot of problems.

I don’t think “better spam detection technology” can help out of this even in theory. The whole point of LLMs is that, by construction, they produce content which is statistically indistinguishable from human text. Almost by definition, any statistical test to distinguish LLM text from human text could be turned around to make better LLMs.

So statistical test won't work, and we need something else

Reply on news.ycombinator.com