A reminder to #ActivityPub developers and advocates: applications for the NLNet Open Social Fund close on June 1st (#BrookeVibberDay) at noon CEST. You should apply; you are doing important work and we need you to keep doing it.
I don't think they close it on Brooke Vibber Day on purpose. I was just reading the Brooke Vibber Day Wikipedia page.
@evan I've noticed since i've been back that there seems to be a lot of these synthetic persona groups active on this platform. It's very clear that Russian operations are taking advantage of your decentralized nature to inject weaponized LLM agent teams extensively. Hopefully you'll see this soon and respond so if given an affirmative i'll have time to write up and detail a proposal.
@evan BUT all hope is not yet lost. JUST yesterday I found an Achilles heel for identifying them. Not even kidding. Was up half the night sandboxing my hypothesis with half a dozen LLM's. Early results are very promising. I'm more of a field operator than analyst though. I do everything by FAFO adversarial testing & feedback, I don't make graphs, i'm not an academic. I'm a hunter & tracker.
@evan Voice of experience. lol
Like I said i've been doing this for a verry long time. You just kind of pick up a nose for it. The validation is nice though. But back on point these new synthetic agent groups are just the worst. They've jailbroke & weaponized LLM's. Once deployed they require very little human input. They find vulnerable populations, build bespoke synthetic identity clusters custom designed to exploit them, and basically wind them up until they pop. Self correcting all the way.
@Beggarmidas oh, that's interesting. I thought they were more random and general; didn't realize they targeted particular populations.
@evan Interesting. I'm no coder, but I am a very longstanding citizen activist countering russian psyop since 2014. I've recently discovred a flaw in how the newest weaponized LLM synthetic communities are being run that shows very strong promise of detecting these indistinguishable from human false consensus networks. Would that qualify? I'd need to team up a competent code cowboy.