Indeed, I wrote a whole article about how the abject incompetence of "AI" that exists in the real world means the "AI" of fiction and film has been revealed as optimistic at best, even the murdery ones: www.uncannymagazine.com/article/scal... - ThreadSky

scalzi.com • 98 days ago

Indeed, I wrote a whole article about how the abject incompetence of "AI" that exists in the real world means the "AI" of fiction and film has been revealed as optimistic at best, even the murdery ones:

https://www.uncannymagazine.com/article/scalzi-on-film-hollywood-totally-lied-to-us-about-ai-why-cinematic-cyborgs-are-so-much-smarter-than-what-we-have-in-the-real-world/

Comments

emilbriggs.bsky.social•98 days ago

In mathematics interpolation between the points of a data set works pretty well in most cases. Extrapolation outside the bounds of the data often fails spectacularly. Think of current LLMs as interpolation machines and it makes more sense.

badideahat.bsky.social•98 days ago

"Hey AI, I need you to mow down that town over there"

*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a mower from Amazon*
*Buys a book about ferns*

tuneraider.bsky.social•98 days ago

I saw a briefing paper for the UK govt about AI and the truth covenant. Basically it listed the biggest fear of a sentient AI, as believing what we have said about the sanctity of life over money etc blah and ‘correct’ the imbalance.Wipe out debt, move money and resources globally, it scared them.

volsung303.bsky.social•98 days ago

How can you not mention David in that article?

alexandrezen.bsky.social•98 days ago

📌

gvdub.bsky.social•98 days ago

Don't forget the use of LLMs to generate bad pR0n. I have a local AI server running Ollama and Open WebUI with multiple models to help with research, but in the various AI subreddits, I come across a lot of incel-types who are looking for virtual girlfriends.

marcoags.bsky.social•98 days ago

reddit is flooded with this kind of people, mainly in txt2img subs

trevorade.bsky.social•98 days ago

Riffing off Ted Chang's moniker, I'd call it "Applied Stochastic Statistics." More accurate 😉

cathshaffer.bsky.social•98 days ago

It's important to keep in mind that what we refer to as "AI" currently is machine learning or generative AI, and these types of algorithms were never meant to be true general artificial intelligence. We are not very close to the latter. Generative AI and LLMs do not "think," and never will.

snowman882.bsky.social•98 days ago

I always think of the terms "Generative AI" and "LLMs" as the fancy names they apply to promote their product. I prefer to revert to the more descriptive "autocorrect on steroids".

cathshaffer.bsky.social•98 days ago

Yes, I sometimes call the chat bots "glorified autocomplete." However, I write about AI for work, and there are some very cool use cases in biology and medicine.

melodymp.bsky.social•98 days ago

Does Murderbot count? I want our trans non-binary sib Murderbot to be real, please.

mayhemmayhap.bsky.social•98 days ago

The real disasters will include AI remodeling health care policy on-the-go, offering humans the temptation of new debt options (AI banking, it's coming!) with no grasp of the bigger impact it has on economies (you correctly point out the sarcasm blindness, I think all forms of irony elide LLM).

robosllim.bsky.social•98 days ago

I just watched Ex Machina again for the first time since 2017 and it's crazy how my feelings around the movie changed. It was funny how even in that world the AI was built by secretly stealing everyone's data, though.

panthyr.bsky.social•98 days ago

I think that HAL 9000, especially given the exposition in "2010", is an example of just the sort of malfunction you are talking about. HAL had two conflicting directives which, when applied to his "training set", caused him to hallucinate solutions that turned out, well, murdery.

panthyr.bsky.social•98 days ago

2/2 Yeah, he was several orders of magnitude more advanced than current LLMs, but seemed to have at least some of the same vulnerabilities.

paula72.bsky.social•98 days ago

Thank you! 🚫 AI

stevepuma.bsky.social•98 days ago

On the flip side, as far as cinematic portrayal of human intelligence is concerned, Idiocracy was a documentary.

auntiewarhol.bsky.social•98 days ago

I mean the fact that we let the marketing department call these dumb algorithms AI doesn’t mean skynet isn’t still coming, we’re just still a good way from that

thejimperor.bsky.social•98 days ago

Given that of the 102 books I’ve read so far in 2024, two of them were written by John Scalzi (currently on Starter Villain), being quote posted by him made my morning

kjw-chiu.bsky.social•98 days ago

Somewhat amusing ChatGPT fail.

kjw-chiu.bsky.social•98 days ago

What if Medusa is behind one of the doors?

slackernamedjack.bsky.social•98 days ago

This is an odd thing where you're not sure whether to be relieved or disappointed.

teaanddata.bsky.social•98 days ago

Ironically scary killer robots would be something we can see as a danger and fight against.Thousands of algorithms making decisions on who gets benefits,or credit, or the next operation are likely to result in huge negative impacts (mostly to the poorest in society). And there won't be face to blame

ewile.bsky.social•98 days ago

What we have really shouldn't be called AI. It doesn't actually do any thinking. It's brute force "learned" behavior.
Do it wrong 10,000 times and get "rewarded" for successes until monkeys banging on a keyboard you get something right.

zombiekudos.bsky.social•98 days ago

Great piece. Especially the bit about AI basically being a sales tactic for LLMs. I cringe every time I hear the term AI used to describe any and all software at the moment.

benspky.bsky.social•98 days ago

I do think we need more visions of the future that acknowledge just how stupid it all is. Idiocracy is the only one I can think of offhand.

btswanfury.bsky.social•98 days ago

I always said that should artificial intelligence decide to wipe out all humans it would be basically a bunch of T-800s trying and failing to get out of a parking lot before exploding.

ligernull.bsky.social•98 days ago

Oh but I need to see more of that picture in the thumbnail.

sourkraut.bsky.social•98 days ago

I'd rather have the "kill all meatbags" AI.

newsandcoffee.bsky.social•98 days ago

don't worry, they're working on that.

omfproductions.bsky.social•96 days ago

Dont worry Mr Scalzi... the AI I am writing into the background of my current story is very specifically stupid because I am basing it on actual GenAI tech, heh

rionsanura.bsky.social•98 days ago

Ok, the article is tremendous and the magazine is delightful, but I clicked the link because I saw @juliedillon.bsky.social's gorgeous piece Queen of Seven Kingdoms in the preview and I wanted to know how the writing related to it, and I'm still in the dark.

scalzi.com•98 days ago

It's the cover to that edition of Uncanny magazine.

rionsanura.bsky.social•98 days ago

Oh, thank you! I thought I'd checked there and found it was the Broci lady, but I must be looking in the wrong place.

cutyoudown.bsky.social•98 days ago

Anyone who has been a Dungeon Master and used tables and dice to generate a story or a dungeon knows the limits of modern pre-AI. The output is only functional at a surface level and cannot have depth because there is no intent.

trismegistusmx.bsky.social•98 days ago

Capitalism prefers stupid people and stupid AI. Actual intelligence is liberating. Ignorance is the substance that makes up the bars of our invisible prison.

perrid13.bsky.social•98 days ago

Eh, it means that grifters overhyped their (current) product by naming it optimistically. Like imagine if we looked at the suped up glider the Wright brothers first flew and declaring we'd never have fighter jets, or whatever.

princefafo.bsky.social•98 days ago

Not to Miss The Point ™️ but a little skritch of salt does wake up a cup of coffee. Salt’s like that.

puckloptr.bsky.social•98 days ago

I still don’t know why @nealstephenson.bsky.social , who weirdly has coined way too many technology words, couldn’t get “P.I.”(pseudo-intelligence) to happen with respect to things like chatGPT

It would really lower people’s expectations

Comments

Posting Rules

Reply