←back to thread

133 points timshell | 2 comments | | HN request time: 0.447s | source
Show context
qoez ◴[] No.44378354[source]
I totally assumed typing cadence and mouse behaviour was incorperated into bot detection for years before this already, interesting.
replies(5): >>44378493 #>>44378607 #>>44378715 #>>44378901 #>>44379408 #
1. lq9AJ8yrfs ◴[] No.44378715[source]
You are not wrong.

The article is more of an intro piece for newcomers and doesn't discuss at all the state of the art or where the competition is--the high end of the market is pretty saturated already but the low end is wide open.

There is a bit of a spread in the market, and the specific detection techniques are ofc proprietary and dynamic. Until you have stewed on it quite a bit, it is reasonable to assume everything you can think of has a- been tried b- is either mainstream or doesn't work well c- what working well means is subtle.

Bots are adversarial and nasty ones play the field. Sources of truth are scarce and expensive to consult, and the costs of false positives are felt acutely by the users and the buyers, vs false negatives are more of a slow burn and a nagging suspicion.

replies(1): >>44379802 #
2. hinkley ◴[] No.44379802[source]
As I understand it detection software is also at great pains to make it difficult for bots to analyze the patterns of rejections to figure out what rule is catching them.

If they can narrow down the possibilities to quadratic space then you lose.