Adi Robertson

blinks in Edward Snowden

One valuable thing in files: investigators' scanned background research (physical newspaper clippings, print-outs of online articles, etc) or email records that included the text of shared online articles.

One notable thing: how many of those article links can't even be found now via Wayback.

A metro station of brutal concrete blocks and jagged black obelisks.

“You finished those DC metro stations?”
“Sure thing boss, beautiful, moodily lit and oddly unsettling, just like you asked.”
“What?”

There are serious obstacles to this working but a publicly funded search engine would be an increasingly valuable good.

the president spending an hour regaling everyone with nonstop graphic murder stories in the style of a preteen who just watched faces of death is a real victory for Everyone Is 12 Theory

Some writers will simply take pics with their phone at home. But few will scale mountains equipped with nothing but piss grenades and determination to get even better shots.

Today, Anthropic CEO Dario Amodei meets with Pete Hegseth at the White House, amid intensifying negotiations over terms that would give the military carte blanche to use Anthropic's AI for lethal autonomous weapons with no human involvement. Heres a look inside those talks.

this all basically makes sense when the defining ethos of the 2020s is to get a high score in any endeavor you take on with the minimum effort required, and only chumps care about what you're giving up along the way

death by phone accidentally turning off dark mode

Smart glasses in court are a privacy nightmare Putting smart glasses on trial.

my favorite subplot of this week's social media trial is "judge makes people stop using meta product for dangerous and harmful purposes while in court to determine whether a whole *other* meta product is dangerous and harmful"

Obviously capability overhang is a thing but so often (especially outside coding) right now it's like, there is no intended capability! It's all overhang!

I'd go even simpler and be like "general-purpose chatbot systems are bad." AI that serves some intentional goal can be evaluated as useful and good or not, a machine with no purpose must be destroyed.

"Pro-human" as a term feels corny but there's just no better way to describe tech-for-tech's-sake fetishism under unfettered control of a handful of fascism-friendly billionaires than anti-human, indifferent or opposed to improving the average person's freedom and quality of life.

I personally wish training/refining open source language models felt more approachable because I would love to try making an AI Dungeon-style interactive fiction piece, but again that's just not something any major commercial system's meant for.

Programming's the main thing I'm thinking about, and ProPublica's done some interesting stuff with it as a search system. I think a tool actually made for/by artists could be a useful force multiplier but in the current environment that's not a great prospect.

Do LLMs, as a technology, have potential value to pro-human causes? Sure, probably. But Grok and ChatGPT as they stand, especially in a landscape with zero regulation or consumer protection, are never gonna get us there.

I absolutely do not want to get into "is the left behind on AI" but I'm unclear on why "catching up" seems to be defined as "use, boost, and fund existing systems" rather than "figure out if the tech's design and ownership structures can be changed to serve pro-human ends"

I'm not going to specify any particular technology here and I would be very careful about weighing this, but policy folks should be a million times more comfortable floating and debating the idea of just banning categories of tech product.

We've held onto the "you gotta train somewhere as the story begins" opening environment, but I see a lot less "click the tutorial button and get dropped into a discrete quasi-demo with a cross-section of stuff you'll encounter in the full game."

It's possible this is in the video but I do think it'd be interesting if we brought back the dedicated "tool around with no consequences" separate tutorial mini-level.

to be fair if you're pushing 40 you were about 18 years old when 4chan was creating all this stuff

I think any reporter or activist using a commercial LLM should probably consider anything they put into it a candidate for being leaked by a hostile tech CEO with the goal of embarrassing them.

I should probably return to it

Yeah I guess that plus several months of early AI Dungeon primed me to think a lot about how the sausage is made

AI Delusions Are Leading to Domestic Abuse, Harassment, and Stalking ChatGPT and other AI chatbots are reinforcing users' delusions about other people — fueling fixations linked to stalking and other abuses.

I am still a little baffled none of this has personally appealed to me because I feel like *exactly* the personality type, and all I've concluded is that maybe I'm too sophisticated in my paranoia to get paranoid delusions from an averaging system.

Jeffrey Epstein might not have created /pol/, but he helped carry out its mission 4chan combined things the sex trafficker loved, like white supremacy and misogyny, and he was close to alt-right figures who drew from it to influence the world stage. 

The original piece is pretty clear that there was never an obvious causal connection, just a lot of cultural overlap in the heyday of 4chan

4chan’s creator says ‘Epstein had nothing to do’ with creating infamous far-right board /pol/ “I regret having ever encountered him at all.”

so @kattenbarge.bsky.social's piece about 4chan earlier this week got an update I didn't see coming: moot briefly reemerging to deny his Epstein meeting had anything to do with founding /pol/

Three panel comic. Panel 1: baby monkeys in overalls jumping on a bed, having fun. Panel 2: one of the monkeys falls over the edge of the bed, reaching out as he falls. Panel 3: one of the other monkeys, now looking super buff and badass, grabs him by the arm to save him heroically. “Monkey Stronger Together.”

godspeed to these folks for keeping pressure on to sort through the post-DOGE wreckage, especially in one of the places its impact is less obvious