LLMs are not intelligent but they ARE good for spouting extruded text that mimics the zeitgeist of their training data and if that training data includes the overflowing sewer that is the unfiltered internet you're going to get gaslighting, lies, conspiracy theories, and malice. Guardrails only prevent some of it from spilling over.
Humans base their norms on their peers' opinions, so LLMs potentially normalize all sorts of horrors in discourse—probably including ones we haven't glimpsed yet.
To be clear, that point about LLMs is a criticism of the training inputs. Which as far as I can see are promiscuously hoovered off the public internet by bots (like the ones perpetually DoS'ing the server my blog runs on this year) with zero thought given to curating the data for accuracy or appropriateness.
They generate perfect "answer shaped objects" … from a foul-minded misogynist white supremacist bigot oozing malice against everyone who's not like him.
yeah, I liked that one too
@markotway @Spoon @cstross The term LLM only describes one specific type, language models (chatbots), and doesn't include the generative "AI" things that produce images or upscale videos, for example. I think the general term that encompasses everything marketed as "AI" is Transformer (or, more precisely, transformer deep learning architecture). All generative AI is bad, but LLMs are pretty much the worst because they don't have any real purpose.
https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture)
@StarkRG @cstross
Also, and very comforting, no-one has yet made any money out of this.
The music has been playing, sometimes clamorous and at times quite softly.
The Tech Bros have all joined in with fervour but it's pretty clear that when the music stops there will be very, very few chairs left to sit upon.
AI is a solution looking for a problem to solve.
What a good post.
You have cleared this up very well for me and I think I understand why the day to day manifestations are so irritating.
I taught on many levels for more than 40 years and plagiarism leaps off the page at me and slaps me across the face.
Any teacher who cannot see AI in student work needs to take a good look at their own knowledge base.
@cstross Bad enough that they hoover an entire site, but what I find incomprehensible is the way that they go back and hoover it again, and again, and again....
Are they expecting the entire site content to change in the intervening few minutes? Or maybe they're just treating it as if it was local storage (and leaving you with the bandwidth bill).
@cstross @delawen The devil is in the details, specially as progress is incremental. Tons of data is being poured into them, even from all kinds of private sources. And reinforcement learning was how chatgpt avoided being just another edgy hate-speech bot that nobody wants to use.
There's so much money in this space right now that, for every criticism raised, there's work already being done to overcome it.
LLMs are the shitbrained version of the ML models that had been accelerating (and transforming entire research areas and industries for the better) until LLMs took up all the oxygen.
To your point, those ML models were built slowly and deliberately using vetted information, which is why they're truly miraculous.
Only this era's humanity would throw away actually revolutionary technology to invest in the image of revolutionary technology instead.
But it's not a pipe, never will be
Eventually they'll realise that ceci n'est pas un profit.
"... those ML models were built slowly and deliberately using vetted information, which is why they're truly miraculous."
There were many wrong turns along the way. A late colleague once gave me a spreadsheet of ML failures. Unfortunately I don't have it any longer, but two failures stuck in my mind ...
1/3
2/3
The ML that was shown photographs of mushrooms and told which were poisonous. Unfortunately the data mostly alternated between poisonous and non-poisonous, so the machine learned that the odd-numbered mushrooms were poisonous.
3/3 The ML was shown photographs of skin lesions and told which of them were cancerous. The machine learned that having a ruler in the photograph indicated a cancerous lesion.
@AnnieG There is a scene in the book where the main character is given the choice between taking the white pill or the black pill.
I went bonkers. But it was a different context. It was the main character's girlfriend letting him chose between a marriage pill and a separation pill.
There a different scene like the red/blue pill scene, but no pills are involved.
@oblomov ah! I had to watch it twice, nice video. I get the link.
MORE touches a different vibe, like, it's "happiness that can be sold to you" and "society made so that there is nothing to do but put happiness in the box". In futurological congress, it's more "why organize society around production, when you can organize it around illusion".
But there is a lot in common.
Also, they tended to be specialized.
"Answer shaped objects" is the perfect description. Filing away for the next time my boss outputs directives based on "AI" search results....
The point about training data is excellent.
I'm waiting for someone to "do it right" - train on a carefully curated selection of vetted public domain or legally acquired texts. But - that's hard to do, so probably a pipe dream.
1/2
The point is that LLM do not think, they have no mind or logic of their own. They just reflect and digest the information they are provided with.
Nor they have an ethical framework or a conscience, and that is what missing here. With humans, we call that an education, taking a considerable amount of your youth and early adulthood.
2/2
For an LLM I would expect an operating model with an accumulated PhD level at psychology, sociology, economy and exact sciences that with arbiter the bullshit it digests before dumping it in the real world.
We do not need psychotic AI nonsense...
And there is a lot of weird, wrong. and sick material out there that is simply dumped on the internet.
Charlie Stross
John Brewer
GinevraCat
Tony Vladusich
StarkRG
Nearly Normal (=>÷)
Mark Otway
Shimmin Beg (he/him)
Rick
klausfiend has moved
Martin Vermeer FCD
Jon Sparks
Ricardo Tavares
María Arias de Reyna
Jef Poskanzer
Niedliche Nacktschnecke ☑️
Steveg58
John
Electropict
Dave
norapap 🏳️⚧️
Annie G
Oblomov
[empty]
Matthew Walton
Bilal Barakat 🍉
Gretyl 
Cavyherd
Tom Bortels
John Timaeus
xs4me2