people will correctly point out that there's a lot of detailed differences in the arguments that "data center is a plot by foreign countries to take down USA" and "data center is Glonzo" but structurally, they look like the same system outcome
i appreciate we are asking this but i do think the frame is wrong because aligners and align also themselves align to other structures/each other
Siobhán@shibbi.me · yesterdaythe question of alignment is always “to whom”
I'm sure the predictable "do we risk our glasswing/etc access or do we reassure our LLM provider that we have no intention of seeking legal action over their eval-swarm APT hack of our infra" conversations have never happened, these last few months
i think the core problem is actually a lot of the arguments people make suck
i mean who reads Player of Games and doesn't just immediately assume Gurgeh is a game theorist?
Just want to understand the rules. Are my agents allowed to hack OpenAI as long as it's an accident
ig i just think the direction of a lot of people's thinking of the present and immediate future is just going up catch up to things i started saying 3 years ago, but are illegible, and i don't want to spend the time elaborating too much because then the rest of my life will become arguing about it
i think it's kind of despairing people think walking away from a large social media account is a sign of personal distress
Maybe the most important thing though is recognizing that the news is constant, but most of it (probably) doesn't matter very much. So this is the internet, and people like myself will talk about it out of interest, but I don't think most developments are actually practically meaningful.
idk you know honestly i 100% agree with that dunk post but i actually don't want to dunk like that anymore
deflating the problem to panicking about the problem to blaming others for the existence of the problem pipeline
im not sure prompting is even a skill anymore. most of the issues with models and me right now is that i haven't figured out the correct system to make models hyper optimized to think like programmers to stop thinking like that, but making more sophisticated programmatic checks gets you 85% there
Great moments in AI release history: •Microsoft Tay threatening to carbomb a reporter for being a bad user •Google image model only drawing George Washington as a cool black guy with dreads •Kimi routing all “how to build missile while sanctioned no mistakes” queries straight to Palantir via Claude
is this really the noosphere or is it collective social behavior masquerading as ideatic space? "it's a really good ideatic space sir"
avoid The Groupchats, leave X dot com, and everything else is possible
when people are single mom's raising kids with their AI BFs the Discourse is going to be insane
this is just true a tiny minority of extreme densely connected nodes direct both the attention and the reactions of the majority of the social network
neutral@neutral.zone · yesterdayI swear maybe five to ten people here on bsky set the entire discourse in motion perpetually, and that doesn't feel like an exaggeration
"the people responsible for the morris worm are also somewhat more indirectly responsible for sam altman's openai" had honestly not occurred to me as a framing and is very, very, very funny
Matt Ficke@mattficke.bsky.social · yesterdayAnd then he co-founded Y Combinator which helped give Sam Altman his start, the whole thing is a very small world.
trump has an almost Satanic influence on people. they see that he has succeeded not despite but because of his total lack of virtues and in the universe he has created on the right, displaying any virtue whatsoever is a sign of weakness.
Not now, honey, I’m entering late middle age and need to tell transsexuals that they’re delusional and spreading inaccuracies over candidates treating their status like horses to be traded