Why Is This Fingerprinting Script Creating Bioweapons?
63 points by haze
63 points by haze
This is a really lovely story. Obviously browser fingerprinting is problematic, but someone made the effort to do something creative and funny.
It is funny to see how keyword based "alignment" breaks down quickly and ends up helping the baddies.
Reminds me of the public service media principle of giving equal space to "both sides". Also known as "5 minutes to jews, 5 minutes to nazis".
Apparently judging intent is undecidable within a single session. They can always iterate on the wording. GEB something I guess.
So we inevitably end up with some kind of long-term reputation tracking. Which itself requires higher amount of trust than we are willing to give to the monitorees, which puts us back at the beginning.
Meta: how should this be tagged in a post-[vibecoding] world? We’ve got [ai] “Developing artificial intelligence, machine learning, LLMs.” and [building-ai-tooling] “The design, ux, development of AI/LLM-powered tools (excludes use).” which don’t feel entirely distinct.
It was submitted with [ai], I suggested -[ai] +[building-ai-tooling], and the immediate result was -[ai] but not +[building-ai-tooling] which confuses me. I’m guessing that (a) it took my suggestion as affirming someone’s -[ai], or (b) it silently failed to add [building-ai-tooling] for some reason like choking on -.