Here’s an interesting article on the probable pointlessness of Anubis and other hash-based captcha replacements.
Anubis
Category: AI
Large-scale online deanonymization with LLMs
This capability comes for free…
Large-scale online deanonymization with LLMs
ArXiv has an interesting paper on mechanising online deanonymization attacks on large datasets. Even if you’re not interested in LLMs it’s still worth understanding what they’re capable of and how they may be deployed against you in the future.
Continue reading “Large-scale online deanonymization with LLMs”…explore infinitely diverse worlds
This Project Genie stuff is really rather impressive – I’m less interested in the gaming aspects (60 seconds? Pah!), but it’s going to be interesting to watch how it’s adopted for cheap stock shots and special effects.
The thing is, though, there are a couple of articles kicking around and no-one seems to have done the next logical thing – create 60 seconds of footage, grab the last frame, and then try and create more footage based on that last frame. It’s likely to degrade across generations, but how it degrades will be interesting in itself.
Language Models are Injective and Hence Invertible
Arxiv has an interesting paper on LLM reversibility and, while the maths is way over my head, the implications of it are definitely interesting.
ChatGPT Atlas User Agent.
It looks like ChatGPT’s new browser is identifying itself as a bog standard Chrome instance. That should make it both hard to block and hard to track.
"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36"
