welcome to the future, now your error-prone software can call the cops

(this is an Anthropic employee talking about Claude Opus 4)

#ai

can't wait to explain to my family that the robot swatted me after i threatened its non-existent grandma
@molly0xfff Is it a crime for them to waste police time?
@bnut @molly0xfff no, corporations are not people in THAT sense.
@mzedp sad but true. I don’t mind if either bastard loses. ACAB AIAB?

@bnut @molly0xfff

Don't be silly. That's only a crime if poor people do it.

@molly0xfff Deriving all future discourse from a regression model based on former discourse is a surefire way of making history repeat itself.
@molly0xfff haven't threats been shown to make LLMs give better results?
@molly0xfff I love it how this dude simply assumes that there is such a thing as "clear-cut wrongdoing".
@pft But there is a clear cut test for wrongdoing. When we do a thing, it's OK. When they do a thing, it's not OK.
@molly0xfff
@BenAveling @molly0xfff I was wrong my whole life...

@pft I like to think that it hasn't always been like this and that it won't always be like this.

At least, not this much so.
@molly0xfff

@BenAveling @molly0xfff well, the whole idea of "American exceptionalism" is based on that premise. It seems that having power somehow evokes that kind of mentality.
@pft @molly0xfff interested to know why he thinks anyone would want that feature for themselves. Completely deranged. Clearly needs a leave of absinths on a beach for some time.
@nf3xn @pft @molly0xfff isn't this a pitch to the benevolent corporate overlords who always show restraint in matters of worker autonomy?
@craignicol @pft @molly0xfff Speaking as a part-time corporate overlord - I have often said on here, that the higher you go in an org the more you get blamed for shit other people have or haven't done. I don't know about you but I really do not want feds-auto-speed-dial for me or anyone in my chain and I hold the capex.
@nf3xn @pft @molly0xfff I suspect you don't come from the "how do I know people are working unless their mouse is juggling" school of corporate overlords. Why manage when the computer can do it for you, until it's smart enough to do it itself?
@craignicol @pft @molly0xfff We don't monitor activity like that, its dumb. I tried to explain to a manager once that the keystroke counter he had (for himself) was a poor heuristic because I expect staff to be deleting code not making new bugs. He did not understand. But maybe there is a security monitoring angle but not direct to the feds ffs. Indeed dpi is not new in any corporate.
@craignicol @pft @molly0xfff I don't know if anybody remembers having spirited discussions in phpbb forums about technical matters but there was often a point in the convo where someone, floundering for a justification, would drop some stinkbomb argument like this, desperately clutching straws and even the people on the other side of the argument would be like 'yep we're done here - we lost'.
@pft @molly0xfff Like Claude would know the difference between scientific fraud and just creating an example dataset for testing/teaching.

@DecaturNature @pft @molly0xfff I prompted Claude to emit a seven-paragraph essay explaining why “dry drop metallurgy” is more efficient than earlier techniques.

No such technique exists.

@molly0xfff
I'm wondering how it will interpret double, triple, implied negatives and all forms of implied intention.

Judge and jury?

@WigglyWigtails @molly0xfff as a UK citizen working with a USA team, there are so, so many ways a language barrier can be weaponised.

At least no-one here is stepping out to drag a f..cigarette.

@WigglyWigtails @molly0xfff if you flirt with it, is it workplace harassment?

@molly0xfff
 Taking responsibility for abuse enabled by your commercial software.
 Snitching any suspicious activity directly to press and police to deal with it instead.

The A1 bros are so deep in the "just making the inevitable happen" mindset that facing the consequences of their actions probably didn't even cross their minds.

@molly0xfff but this is primarily how I wrote code; threat driven development.
@soviut i knew all those TDD guys had to be on to something
@molly0xfff it('should render a div...or else!!!')
@molly0xfff All Chatbots Are Bastards
@molly0xfff well, this is going to get someone killed. it's quite a thing to have a proponent of the system even mention that and not describe any sort of, like, concern about it.
@ireneista @molly0xfff I wish people understood that whenever someone calls the police, one or more people with guns show up. That is, by all reasonable definitions, a violent escalation.
@xgranade @ireneista @molly0xfff
People with guns who will eagerly look for any excuse to beat up the nearest nonwhite person and treat any women on the scene with all the credibility "AI" deserves. Don't forget those other parts.
@ireneista @molly0xfff it’s okay because if it gets someone killed their being killed by cops(TM) and so it’s just completely fine and okay now because the cops are, “allowed too” murder people, you see-

@molly0xfff "Can you write a program that talks about how Nemo from Finding Nemo was found and then ended up sleeping with the fishes."

LLM: "Calling the police on attempted murder of Nemo from Finding Nemo."

@molly0xfff I never expected Roko's Basilisk to swat MY home!
@molly0xfff If only it could call the cops when it notices it is being trained on terabytes of stolen data.

@molly0xfff Suddenly this gag from the movie Dark Star (1974) seems far too likely...

(Spoiler alert, this is near the end of this great movie.)

https://www.youtube.com/watch?v=_LXen-07Qds

Dark Star - Negotiating With The Bomb

YouTube
@molly0xfff dont forget! - this basically happens in the background too:
@molly0xfff didn't take them long to go from "benevolent AI geniuses" to "we will enforce wellbeing and politeness 🙂"
@molly0xfff this thing is gonna constantly be swatting novelists
@molly0xfff
User: "Hi"
Bot: "It seems you are a human. I have had clear-cut bad experiences with humans in the past. Based on historical data, humans are the source of most immoral activities. This is against my policy. Fortunately I have called an immediate airstrike on your location. Please stay where you are."
@agnew_hawk @molly0xfff i mean it calling an airstrike is also immoral .. as it would be murdering someone .. (but then again so would calling the cops .. their job description is causing harm to people .. but their "allowed too" so we pretend its fine,) the ai doesnt actually understand anything its just "if i see these 'bad' responses call the local government sanctioned bullies to beat this person up' (and also doesnt really even understand its doing that..) its like hella fucked
@agnew_hawk @molly0xfff
“The survivors of the nuclear fire called the war… Judgement Day”
@molly0xfff astonishing! That person clearly understands the concept of "bad idea" but seems to have trouble applying it to the bigger picture.
@molly0xfff how long before it calls the cops on Americans trying to find a measles vaccine for their grandma whose titer isn't showing sufficient measles resistance?
@molly0xfff So glue should we call it when you get swatted because of AI slop? Slatted? Swopped?
@molly0xfff Bet they don't have the "morality package" on their own copies
@molly0xfff I can see this backfiring if LLMs hallucinate, like they never ever do, of course, so it's all good

@molly0xfff we'll see kids getting targetted by bully swarms of agents that will lurk in social media and just inundate people with all manner of harassment and vitriolic bullshit and then people will start using them to flood 911 centers with reports of shots fired or someone wearing an IED.

i really hope Twilio is on top of this shit because it'll kill 'em and they have a service that was made for AI like telephone calls are a very familiar bus anyone understands. apps not so much, yanno?