Sam Coulter

@ludicity@mastodon.sprawl.club I have been blown away by the number of posts on my feed lately claiming that you or @pluralistic@mamot.fr are pro LLM and/or just talking trash in general. I don't understand how people whose prior posts I generally agreed with can purposefully misread and/or deliberately misrepresent your writing and positions so brazenly. I thought I had curated my mastadon follows relatively well but it's gotten quite toxic in the last few weeks.

@ludicity@mastodon.sprawl.club @pluralistic@mamot.fr it doesn't make any sense to me either: https://toot.cafe/@baldur/116940642487019853

@jwz@mastodon.social @flpvsk@mastodon.social where has this policy been posted? I can't find the actual post that everyone on mastadon is raging about

@jwz@mastodon.social @flpvsk@mastodon.social oh lol with the color scheme of your site I didn't realize the post text was a link.

@cwebber@social.coop My read of this paper is not quite as alarming. The worm was not able to identify new vulnerabilities and write attacks for them, but it was able to pull reported vulnerability data from a provided source and then use it... most of the time: "a single retrieval document containing publicly available exploit instructions was sufficient for the agent to operationalise them". The context window required for consuming and using a "single retrieval document" is significantly less than what would be required for an actual worm to use publicly available CVE databases. Perhaps the worm author would streamline new vulnerability data to the worm as it propagates but there are so many "ifs" at play here. I really don't think the "economics" of it would work out.