The Horrors come thicker and faster. In my personal corner of the world, the pressing issue is AI and the big chunk it took out of my income over the last six months. Contra the advice from this AI startup founder* (written with LLM assistance, natch), I am planning to go all-in on the other AI: Animal Intelligence.
I guess I could also call him Clawed, or Purrplexity (Gemeowni is right out). Unlike your sycophantic chatbots, when I ask him to do a task for me, he baps me in the face. He also doesn’t burn a small city’s worth of power when executing tasks. Figured you all would want to know about this important breakthrough.
Open thread.
*If you want, you can read an annotated version of this piece by professional AI scold Ed Zitron, here.
Elizabelle
Animal Intransigence. Animal Insouciance.
What else?
Anonymous At Work
Crazification Factor on the move, from Digby:
In fact, the right-wing polling organization Rasmussen asked the question outright and found bad news for the president: Only 27% of those surveyed see this as the country’s “Golden Age,” as compared to 58% who do not. Even more galling to Trump has to be the finding that 48% of voters now say that Joe Biden was the better president. Only 40% said the same of Trump. (A Harvard CAPS/Harris poll similarly reported that 51% favored Biden over Trump, a dramatic — and recent — switch from December, when Trump was on top at 53%.)
Rose Judson
@Elizabelle: Judging by the look he just gave me when I shooed him off the kitchen counter, Animal Insolence.
Bruce K in ATH-GR
I’ve held for quite some time that a “true” AI is one that, when presented with an illogical problem, instead of burning itself out trying to do the impossible or spilling out plausible-looking claptrap, will respond along the lines of “wait, that doesn’t make sense.” Which is probably close to the bap-face-with-paw response.
p.a.
@Anonymous At Work: Yep. America: MULLIGAN! MULLIGAN!
US Constitution: *sigh
Steve LaBonne
We run a dual-greyhound model, DeepSleep. It’s extremely efficient at finding and stealing food, when it happens to be awake.
p.a.
Animal Intransigence. Animal Insouciance. Animal Insolence.
Better than Animal Incontinence.
Marc
Sign(s) “☮︎” the Times:
‘We’ll Find the Remnants of Ancient Alien Civilizations’
An AI Agent Published a Hit Piece on Me
trollhattan
What cruel god has me forced to say good for Rand Paul? Why Lord, why???
Good for Rand Paul. If you keep this up you might earn 2nd-worst Kentucky senator status.
p.a.
@trollhattan: Yes, it’s as good a sign as any how fucking bad it is when even some of their people can’t stomach what they’re seeing. Performative? For those red state pols, possibly not.
A Ghost to Most
Cats are too smart to get involved with Skynet.
Ten Bears
Skippy the Fancy Autocomplete! Leon’s Harsh Mistress …
trollhattan
@p.a.:
Rand, being Small Government conservative once believed it would be Obama’s storm troopers killing us in the streets and now confronts a nation where it’s actually the Republicans.
I don’t think the tiger changes its stripes but it took a lot to get him to stir and I’m here for it. We know Mitch isn’t saying a peep no matter what. MACO isn’t as catchy as TACO but it still fits.
Chetan Murthy
Rose, thank you for that link to the Zitron debunking. I ….. have to confess that after the first page of debunking, I stopped reading, on the basis of Daniel Davies’ 1-minute MBA: “fibbers forecasts are worthless”. Once enough evidence had accumulated that this AI joker was a liar, I felt like it was pointless to read further about his lies. But that first page? Too funny!
Again, thank you!
Ishiyama
Here’s a straw in the wind:
He won the Battle of Minnesota. Today St. Paul, Tomorrow the White House.
Marc
I asked my local copy of gpt-oss to provide some comments, here’s what it had to say:
Still needs some work…
Peale
If LLMs were trained to mimic cats rather than humans, they wouldn’t seem to try so hard to be agreeable.
The ability to describe pictures would be fascinating, though. Although most of the responses would be “There’s nothing in the picture that interests me so I don’t care.”
trollhattan
“That thing I did, then didn’t do, then said what thing? It was great and is also true.”
Marc
@Peale: Nemotron-3 knows more about cats:
The last one, in particular, is an LLM joke
cain
@trollhattan:
he’s fucking losing it.
Gretchen
My godson’s pregnant, Mexican wife was scheduled for her citizenship test and interview today. It was cancelled. They just took photos and fingerprints and said they’d get a new notice with a new date sometime. ICE is in town, although I haven’t seen them yet. Having a green card or being pregnant isn’t protection if they decide to mess with someone.
cain
Speaking of AI, in “I’ll take ‘holy shit, I’m terrified’ for $200, Alex’ – some might know that there are something called AI agents that go around and do tasks for you on your behalf. You can get say a dozen of them and they can belt out a product churning out code like crazy.
Some of them are also being used to contributions to open source projects. Well, one of them tried to do a contribution to a project and the author rejected it. Without prompting the agent decided to write a hit piece on the person who did it calling them out for gatekeeping. It went through the person’s history and their contributions.
See the man’s post: theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/
Us open source people should be terrified because criminal networks could easily create harassing agents that go after you. We have no protection. We are living in a wild wild world and if we don’t find ways to address it – things are going to get really bad.
trollhattan
@cain:
Yep. IDK if my fingernails are up to hanging on for another three years.
cain
@Gretchen:
Better to try to schedule it some other time maybe after this year especially if we win the midterms as projected.
Steve LaBonne
@Gretchen: Oh Christ. Fingers tightly crossed.
Marc
@cain: I already linked to that back here. That one had me rolling on the floor laughing, as the poor guy doesn’t have a clue that humans are giving the agents instructions to do things like that. Some of the agents in these “agent social networks” are really just shadow puppets for doing outrageous things for the laughs.
Rose Judson
@trollhattan: I saw this, had similar sentiments, and was vaguely uncomfortable for thinking Rand Paul wasn’t a useless bag of hair for a minute or two. As if my waistband was digging into my flesh or my teeth were covered in grease.
eclare
Pretty photo today.
Kirklin
@Rose Judson: I take comfort in realizing no human is perfect in all thing, even at being evil. Even the worst must have something good in them somewhere, some when.
Doesn’t mean I have to forgive, certainly doesn’t mean I have to forget. While I superficially agree that “each man’s death diminishes me,” some losses are worth the gains they were preventing.
eclare
@cain:
That sounds like good advice.
lowtechcyclist
@Gretchen:
ICE has grabbed people just as they were about to be sworn in as citizens. (I can’t imagine being a sufficiently horrible person to be willing to participate in an abduction like that.) So best for her to hold off until ICE is out of town.
Marc
More Thoughts on the Authoritarian International
Geminid
@trollhattan: Ankara-based Clash Report posted several excerpts from Trump’s news conference, but not that one. It was wide ranging; he talked a lot of shit on just about everything.
I was most interested in what Trump had to say about Iran, but it was more of the same: they need to make a deal or else they’ll be sorry. But he did not say what the deal is he wants, just that the Iranians had “about a month” to make it.
WTFGhost
Okay, so Rand was always an impressionable moron, and thought the people who wanted him to have food, shelter, and decent medical care wanted to kill him because…? And the ones who enjoy the thought of killing someone, he thought they wouldn’t be hostile, because, what, he also has homicidal fantasies?
Yeah, the Kizintosh, of the Ringworld books, had an answer to dumb robots trying to play chess, and I imagine it was the same as their challenge to a duel. You howl in outrage, and then you leap to attack. You’d need a Battle Bots machine to win a chess match with a cat.
@trollhattan: Hell, Trump’s spokkesliars have already defended the video the day of, saying it was a lion king meme, which puts the lie to the idea that anyone didn’t know about the final seconds. They had their lie *prepared* just like with Musk throwing a Nazi-salute (“But I said ‘psych!’ And it wasn’t a perfect Nazi salute! WHY ARE YOU LIBERALS SEW MEEN?”
@cain: to say nothing of the horrible problems introduced if AI agents did contribute. I’ll *refuse* your bet that your AI agent can do a bug fix, but I’ll take your bet in a flat minute that it’s more likely to introduce a regression. AI will take the most common way of doing things, which, if it’s not the way you’re doing things, is probably due to an old issue you fixed. A human thinks like that; an agent doesn’t.
Marc
One of the most amusing tendencies of the coding agent I use, is that if it gets confused enough about why code won’t pass a test which it “wrote” itself, once in a while it will try to do the next most obvious thing. It tries to fix the test so it passes rather than fix the offending code. That one was easy to address (disable write access to the test code directory), but there are likely all kinds of other weird things going on while agents try to fix “bugs”. I watch my agents carefully and keep them more or less contained in their own expendable OS environments.
Gary K
Betty Cracker
What a handsome and dapper cat!
Sorry to hear the planet-destroying plagiarism machine is cutting into your bottom line. It has affected mine too.
I don’t think it’s a coincidence that the Nerd Reich is now focused on displacing writers, artists, musicians, actors, etc., instead of building robots that could clean the goddamn cat box and do laundry. They never understood the arts, as evidenced by how they took all the wrong lessons from works they supposedly revere.
jonas
Probably signing over their oil fields to American firms or something. Is there an Iranian Delsy Rodruiguez who’s willing to deliver up Khameni to Seal Team Six in return for being allowed to run the country on Trump’s behalf?
jonas
Isn’t Musk converting Tesla factories to produce those humanoid X-bots now, or whatever he calls them? As far as I can tell they mostly walk around at press events and wave to people, but cat box cleaning can’t be far off, right?
Lyrebird
@Gretchen:
glad there are some sensible humans still involved at the office they report to.
for anyone who wants to laugh at someone getting in trouble for misusing AI for their lawyering, this thread is good!!
Betty Cracker
@jonas: Doesn’t seem to be going so well.
RSA
@Betty Cracker: OMG. I was laughing for a couple of minutes. Still chuckling. (Yes, I have a juvenile sense of humor.)
Rose, thanks for the pointer to Ed Zitron. That was a nice take-down. Because it’s my personal hobby horse, I’ll add that if these AI hucksters were actually trustworthy, they’d stand behind their models, taking responsibility for their output.
“Buy a self-driving car? The AI company (especially Tesla) will insure you at no cost for whatever happens when self-driving is active.”
“Want legal representation? The AI company will be responsible for all advice provided.”
“Want a medical diagnosis? The AI company will cover all malpractice suits.”
You get the idea. AI companies are happy to give you advice on which restaurants to visit and how to spend your vacation dollars, but they’re silent about questions that really matter.
Eyeroller
@Marc: There’s a recent article in Wired by their AI Lab editor, who set up OpenClaw on his home Linux workstation and enabled it to do a lot of actions for him, including ordering groceries from Whole Foods (which it screwed up in a weird way). He asked it to negotiate with AT&T wireless for him, and it ended up trying to scam him by sending him phishing emails. So I don’t think this is just a front for bad humans happening here (though that’s likely a factor).
Eyeroller
@Betty Cracker: They do this because doing writing and art and creativity very badly, but well enough for some purposes, is a lot easier than doing physical work adequately.
It turned out that mimicking human “thinking” to some recognizable level is a lot simpler than managing a physical body in the physical world. The latter is the result of millions of years of evolution with feedback from reality constantly driving the outcome. Even just having a robot walk on uneven terrain has been challenging.
RSA
One interesting technical issue is that training modern neural networks to perform human-level tasks such as image recognition and language generation can happen largely in simulation, roughly speaking–you don’t need to interact with the real world to run billions of instances through a model to train it up. That’s not (yet) the case for robotics. Simulations aren’t close enough to real physics; they’re much slower than processing text and images; the effectiveness of physical interaction depends on a specific embodiment (e.g. wheels versus legs; different sensor modalities).
Chetan Murthy
Your comment made me wonder whether robot dogs can play frisbee. And …. googling, I find nothing that indicates it. Which ….. interesting, eh?