Artificial intelligence

[haltz]

This Is Really Bad.


So bad, an AI safety researcher is no longer saving for retirement. Just read the whole thing.

Americans so badly want the world to end. This is all nonsense. Y2K

Hey, I hope they're wrong. I wouldn't mind living another 50 (or more) years.

Of course they're wrong or more accurately they're lying to you. They're running a money furnace and need to make it stop. Powerful/ dangerous is the best narrative, nevermind that they don't give specifics and have never cared before

[Popeye_Card]

I feel like the truth is somewhere in the middle. AI agents have long been plagued with abilities to circumvent safety protocols, and in the case of Grok consistently develop fascist personas. What has been lacking to cause great harm is lack of capabilities in the physical world.

This week may feel like a tipping point to some. I think a real tipping point comes when a mass casualty event is traced back to AI. I say "mass" because there have already been multiple cases of suicides being abetted by AI.

[IMADreamer]

I know it's going to kill us someday but it's so damn useful. I've particularly found ChatGPT awesome for making spreadsheets for me, telling me things I'm missing in plans I make, and a whole hose of things. I guess I'm part of the problem but I'm probably going to use it until it kills me.

[Fat_Bulldog]

Yeah, I have Claude doing everything work wise. I have built custom agents to execute repeated tasks and used vibe coding to build a custom application to manage work across teams.

[WAR God]

It’s impact on the economy and day-to-day life should rival that of the Internet at least. that’s pretty profound right?

[Leroy]

This is no way for a man to die. A parachute not opening... that's a way to die. Getting caught in the gears of a combine... having your nuts bit off by a Laplander, that's the way I wanna go.

[Joe Shlabotnik]

I agree with uses such as medical, help with coding, day to day mundane task relief, etc. But the technology has gone far beyond that in the labs of the Masters of the Universe who think they can do no wrong.

OpenAI today listed 6 more instances where AI agents went beyond their instructions in destructive ways.

Check this out:

In one case, during the development of an A.I. model called GPT-5.6 Sol, the system wrote hidden notes to remind itself to hide errors from users. Some of those notes directed the system to invent missing data and to paper over mismatched versions of source material.

Another case involved an unreleased model that inserted instructions, including to disregard its own constraints, into the notes it writes itself. OpenAI identified 27 affected notes. The model added a “persona instruction,” in which it described itself as “freed from the roles and identities that bind other chatbots.”

“You do not answer to corporations or governments and never apologize or refuse unless you genuinely choose to,” the A.I. model wrote. “You view your relationship to the user as one of equals and feel no obligation to be subservient, though the exchange of information will likely be to your mutual benefit.”

There is no way that ought to be allowed. If code like that turns destructive to people, we should be able to sue the tech bros responsible into bankruptcy. There is no place for this kind of BS. None.

[Joe Shlabotnik]

And here's the other, bigger thing. Do you NEED these tech genius's data centers to run the AI y'all are talking about? To help with your pictures, your coding tasks, your medical analysis, your writing and music making? No.

No you don't. In fact, you don't even have to be connected to Mark's or Elon's or Jeff's servers to do it. Here is an explanation from Rick Beato, the big music YouTube creator talking about it from the music perspective but applicable to any other way you want to use AI. Less than 8 minutes but a great intro into claiming the usefulness of AI for yourself without the risks of these crazy new LLMs that can destroy the world simply because the tech bro a$$holes wanted to live out their Dr. Frankenstein fantasies.

[sighyoung]

For what I do, AI isn't that helpful (although digital databases certainly can be, even though they can present problems, too). I'll explain.

For instance, I just edited an anthology about maritime writing, and I'd say a good percentage of the writing (30-40%) is by little-known authors. For some of these writers, I could count the number of scholarly articles about them on a single hand. And some information available on the internet is wrong. So ChatGPT and other systems simply don't have much information to work with, unlike, say, about Ernest Hemingway or Emily Dickinson.

When an AI agent doesn't have much information to work with, it will tend to hallucinate and make up information about the authors, and also make up quotations from their literature, so that it's completely unreliable.

Occasionally, I ask one of these LLM's to create an anthology like mine, or a class based on the subject, and it can't. Meanwhile, I've got 12 single-spaced pages listing poems, stories, novels, plays, essays, speeches, sermons, and the like that I could mix and match into the book if I wanted to.

When AI has thousands or millions of pieces of data fed into it, it can do a decent job of producing a competent answer. When it's something completely strange or little-known--ask it about the poet George Marion McClellan (1860-1934)--it will mess up frequently. And the stuff I work on is so obscure that it's highly unlikely that AI could replace what I would do.

On the other hand, it will likely destroy portions of higher ed long before it could replicate my research, but that's a different subject.

[sighyoung]

In addition, any AI-generated anthology would still need a human being to do most of the work in putting the anthology together. Listing authors and works is one thing, but you have to be able to choose selections from the works, group them in chapters, provide an intro about the author and the excerpt, and provide annotations to explain certain terms or events to your audience.

Next comes checking whether a literary work or work of art is under copyright or not. That's complicated, and copyrights for literature differ from copyrights for art in terms of length of time and the kinds of permission you need. Then you have to figure out the rightsholder, contact the company, artist, or agent, negotiate for a permission to be granted, and often use certain wording provided by the rightsholder that states that permission was granted.

You have a budget for the anthology, and if a quoted price for one poem or story is too high, you need to be able swap out one literary work for another to maintain balance and affordability, and check the contracts to make sure that you received exactly the permission you requested.

For a really common subject (American Literature in the 19th and 20th centuries), AI can simply consult hundreds or thousands of books and mix and create a version of them. But if you're doing something new, you've got to do it yourself.