John Scalzi: One of my favorite things to do, paraphrased: Me (to "AI"): Recount the plot to [one of my books] and name five major characters and their roles. AI: [Generates bad plot, correctly names only the protagonist, all other characters and their roles made up] Me: That's wrong, you know. AI: I know.

One of my favorite things to do, paraphrased: Me (to "AI"): Recount the plot to [one of my books] and name five major characters and their roles. AI: [Generates bad plot, correctly names only the protagonist, all other characters and their roles made up] Me: That's wrong, you know. AI: I know.

A lot of people don’t understand that Generative AIs will just make things up. They’re more like Mad Libs than what we used to think of when we said “AI.”

Oh, God, I just did this with one of my own novels and the results are hilarious. The novel is on-line and free! It got, like, 30% of the plot. It actually identified the theme accurately, but completely hallucinated the three central protagonists!

Mine was pretty close and I'm wondering if that's because it was in the stolen data set used to train AI.

I tried this with both Star Kingdom (on both AO3 and FF-dot-net) and TftP: Honest Impulses (an original work), which has been on my website and a number of others for several years now. The AO3 one only identified half the fandoms and got the characters wrong. The OW was terribly wrong.

I mean, kudos for it understanding that Star Kingdom was a Frozen fanfic, but it missed the entire Honor Harrington side of the story. (It's weird; I wrote it on a drunken bet and am surprisingly proud of it.) But it just made a ton of shit up, and that's... not very useful.

Hang on…. Frozen and Honor Harrington??? Off to find this immediately ….

It's a lot more Frozen than Manticore. I was with a few writer friends one night and got ranty about how Weber's economics Make No Sense. There should be zero incidental poverty in a universe with cheap lift-to-orbit, so one said, "I bet you could write that." archiveofourown.org/works/454266...

The Star Kingdom of Arendelle - Chapter 1 - elfsternberg - Frozen (2013) [Archive of Our Own] archiveofourown.org An Archive of Our Own, a project of the Organization for Transformative Works

It's why I hate using "AI" to describe these algorithms. They aren't AI, they're autocomplete with delusions of grandeur.

Back in the IRC days of NaNoWriMo, the guy who ran it had a chatbot program named Timmy Timmy pulled random words or phrases from the chatters and would chat with us and also write his own 50k novels Everything he said was utter random nonsense that would randomly make sense TIMMY was a proto-AI

So when AI started becoming a thing, the IRC gang still remaining went to each other and conferred "This is just Timmy, right?" "Yeah, it's just Timmy but with more programming." So none of us fell for the "AI is amazing!" nonsense because ALL OF US KNEW TIMMY And Timmy was just a goofy program

AI is just a goofy program that too many people are taking seriously. It's ruining careers, it's ruining lives, it's DRAINING THE POWER GRID Timmy eventually got shut down and "died" We need to do the same with a lot of this AI stuff

I'm not going to claim it doesn't have its applications in science and medicine, but for creative? It's just TIMMY Annoying, bizarre, occasionally making a weird sense TIMMY And Timmy did not make good art or books They were unreadable AI needs to stop being Timmy and go into science

one very clarifying thing to realize is that AI (really LLMs) do not give you an answer to your questions. they can only give you words in the shape of an answer.

That’s how I graduated college.

Creative plagiarism. Heck, I even plagiarized myself once.

It doesn’t even know! It just has analyzed “I know” as a phrase that reasonably follows “you are wrong”

It needs to use emojis to show it's really ashamed

Once asked it to tell me the ‘scary’ parts of some Pixar movies. There were scenes in Cars that I very much missed…because they were fully hallucinated. 😂

Oh, wow. This is a hoot. It's the confidence with which it is so very wrong that's impressive. Honestly though, the danger here with my novels is that there is not a single word about them being erotica. And... Oh dear. It says one of my novels is "a children's fantasy novel." Uh yikes. No.

My Google phone gives the wrong answer to simple questions like "what station is the _____ baseball game on". Used to always work.

So, is that a better or worse success/accuracy rate than your average human reader? As I've pointed out to my hot-to-use-AI boss, you'll end up wasting *at least* as much time training and verifying the output as you will with your more problematic employees.

The sad thing is tech bros would probably tell you that clearly you wrote the wrong book and should revise with the AI version

Ok but please stop asking them to use more resources? Some of us have anxiety

Does it say "I know" in Han Solo's voice, though?

I asked ChatGPT for a summary of Cards on the Table by Agatha Christie. It did a great job...up to but not including the very important final twist.

Should have been called Regurgitative AI.

Should have just named these systems “George Santos”

Yes, and the problem is that people do not distinguish an answer (what we want) from a text that *looks like* an answer (what an LLM can provide)

Like shit, I can do that without using megawatts of power and, ok I still use gallons of water BUT THAT IS A MEDICAL CONDITION

And then it smiles

Future Elon Mars Mission: Elon: "AI, is the air outside breathable?" Ai: "Yes, the air outside is breathable!" Elon: *Opens the hatch"... End of mission.

It knows _now_

No, it doesn't. If he asks it again it will produce similar information-shaped sentences.

Well, at least it was honest. . . .

Post