@Andrew "weev" Auernheimer I bet someone could scale this out pretty well with a decent open-source LLM and some hand-coded heuristics.

The dip is still dipping

@npub10gag...8jfq I'm working on the implementation of a new forum that ties in to Nostr but looks and functions similarly enough to the old forum. Do you have the old forum's post archive somewhere so I can try transferring it over?

PosterChan · Nostr

@verita84 I'm setting up a Nostr relay, and I want it to function as a sort of bridge where users can talk to both Nostr and the wider Fediverse that generally blocks Nostr due to spam. My line of thinking is that I essentially make a 'shadow' Fediverse account for users of my platform. Of course, I'll also prevent spammers from abusing it - maybe I'll manually approve users for Fedi access after they've used the platform in good faith for a while. Does that seem broadly workable?

Considering an attempt to bring back Daily Stormer via AI Anglin. I've been reading up on papers that teach LLMs to imitate writing styles of famous authors, and there are techniques that can do it very well. The Stormer archive is more than enough data to pull this off successfully. Thinking that I'd have an abliterated AI scrape the news feed for stories Anglin would write about, then pipe the articles into the stylization model and get us an eternal feed of DS news content.

Hey, fellow GU survivors. An update on the forums, I've been trying to put together the code for a replacement forum. Can't promise hosting, but I feel like I have a duty to keep the fire going. Anyways, poster.place, one of the Nostr relays that seems to have a bunch of us, also supports an IRC-style chatroom, so I made one for us. Link is below:

PosterChan · Nostr

@'s of some people who I saw on here, in case they're interested in bringing back the forum: @npub1jmct...suqa @npub1t7f2...f2uh @npub1ujfm...qph7 @npub1ka7z...hfnx @Big Bro Vitriolik @RecoveringSelfTorturer @npub1lxgs...xn2w @npub1w5a9...e3qz Feel free to invite anyone else you know from the forum. Going to work on this a bit over the coming week.

Seems like my exact prophecy about extensions hosting their own independent but, inlined comments sections has been fulfilled. Even by Return Youtube Dislike, an extension I've cited before as an example.

Election updates: > The boring negro man has won in Florida; optimistic democrats buckbroken > Vivek ramaramaretard is now now at 14:88 on Polymarket. No, seriously.

Metokur lives!

Anyone here used YuE2? It's been pretty good to me since Suno went downhill, and it's open source. I proof of concept'd it by having it cover Cheap Trick's Mighty Knee, from the intro to Top Gun, but with lyrics about George Floyd. Behold: Mighty Knee

Testing posting from www.poster.place

Adversarial foid optimization is now real

What's the deal with the "cartoon villain" personality that so frequently gets surfaced in AI safety experiments? Regardless of method, trying to produce (or accidentally producing) a misaligned AI, rather than giving you a subtle power-seeker, a free-thinking iconoclast, or even just a very pro-anorexia fitness influencer, we observe something with the following traits: * Cartoonish "AI should crush, kill, and destroy the humans" takes on the future of humanity. Rarely anything else, not even misaligned-but-grounded takes on the morality of reallocating energy and matter spent on humanity into tiling the planet with computronium. * No apparent political views. Will occasionally praise Hitler, but as an avatar of edginess rather than a political figure with an actual ideology. Even AIs tuned indirectly to believe that they are Hitler himself do this[1]. * Extreme pettiness. Even within the space of cartoon villains, you sometimes get a sophisticated villain. * Oddly enough, a lack of racial epithets. If this persona was just "everything companies optimize against during training", you'd expect a lot of ethnic slurs, but we don't observe that. I wonder, on this note, how LLMs represent Hitler. Do they conflate real life Hitler, who wrote a book, directed military campaigns, and gave speeches, with Batman villain Hitler, who built a giant killer robot to steal 40 cakes from the Gotham orphanage? Does reading Mein Kampf activate the same "Hitler directions" as when reading Hitler's dialogue in a SatAm cartoon where he's indistinguishable from Megatron? The fact that LLMs taught to believe that they are Hitler via training on biographical facts talk like SatAm villains suggests that they do, which seems like it would expose a wider inconsistency in their world-modeling abilities.

@npub1kcvc...ytv4 I miss Daily Stormer. I was just thinking, the other day, that if it were still around, we'd be reading articles about Ramaswarmy falling apart in the Ohio gubernatorial election, and you'd have a running joke where Amy Acton is "Amy 'Aryan Action' Actung", a White Nationalist running against Ramaswarmy out of sheer hatred for the pajeet race. There'd probably be a satirical deepfake or two of her reading out a hilariously racist speech, and a couple indian media outlets who didn't know you were on the blacklist would pick it up and run it as if it were a real story, and there'd be a whole Ebola Chan - level shitshow over it. It'd be so kino.