Post

Karl Bode
@karlbode.com

honestly some of the most apocalyptic impacts of "AI" will be shit like this, and most of the doomsday dipshits predicting SkyNet can't be bothered to notice or care

arstechnica.comUnitedHealth uses AI model with 90% error rate to deny care, lawsuit allegesFor the largest health insurer in the US, AI's error rate is like a feature, not a bug.
Nicholas Grossman@nicholasgrossman.bsky.social

There have always been humans who thought: 1) The world will end soon 2) It will happen in a cool, exciting way 3) It’ll happen to me personally 4) I’ll be a martyred prophet, or a warlord, or otherwise special 5) People who don’t see this are lesser “Sexy apocalypse” is a great phrase for that

Karl Bode
@karlbode.com

and certainly our shit corporate press would rather make ad engagement clicks off of bullshit doomsday headlines than engage in nuanced stories about the ugly collision between corruption, automation, and regulatory capture

57 reposts423 likes

pietro spina
@pietrospina.bsky.social

been telling reporter pals to watch for the head on collision between disintegrating rural healthcare and ai. the story will come twice. first as savior, then as failure.

Nicholas Grossman
@nicholasgrossman.bsky.social

Your post clarified something for me. When I wrote about LLMs in military targeting, a striking detail was people who followed or rubber stamped the outputs feeling absolved of responsibility. They didn't choose who to kill, they just followed the machine. Rejecting insurance claims fits that too.

Nicholas Grossman@nicholasgrossman.bsky.social

For example, LLMs are involved in military target selection. Right now. And some humans in the loop give AI recommendations a rubber stamp, expressing relief that they feel less ethically responsible. But worrying about that and figuring out how to address it is hard. "What if Ultron?" is easy.

foreignpolicy.comKeep Humans in the LoopAI has a place in military targeting—but it needs safeguards.
Karl Bode
@karlbode.com

Which is why it would be nice if the press could quickly steel itself to not using the industry's chosen nomenclature for a lot of this stuff, like "model misalignment." It's all very intentionally depersonalizing and distancing from responsibility

Mighty Apollo
@mightyapollo.bsky.social

I see it in LLM coding use as well. The users seem to get a dopamine hit from immediate gratification, but also have an imaginary friend to pass blame onto. I’ve now heard “Claude must have…” excuses more times than I can count.

Correct Time (CT) Zone Guy
@shmody.bsky.social

How would top diplomats, hostage negotiators, war crimes lawyers, primary school teachers, Nannies of the most rambunctious children, organized crime leaders, or regional community leaders negotiate with a rogue / misunderstood AI agent(s)? #Negotiation #Mediation #ArtificialIntelligence

A_lethe_ion
@aletheion.bsky.social

AI does the one thing that really helps perpetrators of mass violence... okay two things it takes responsibility away from doing the horrible shit explicitly reinforces the frame of reference where killing for personal gain is being artificially lifted up as heroic, but the usa is expert in that