So the third thing I asked Chat GPT to do -- as discussed in the immediately prior post -- was to write a short story in the manner of Robert E. Howard. Here is what it came up with.
The sun beat down on the dusty plains as Elric the barbarian rode towards the small village. He had been wandering for days, searching for a way to ease the restless fire that burned within him.
As he rode into the village, he could sense the tension in the air. The people were on edge, their eyes darting nervously at him as he passed by.
Ignoring their wary glances, Elric strode into the local inn, his hand resting on the hilt of his trusty broadsword. He knew he needed a stiff drink and a warm meal before he could continue his quest.
The innkeeper eyed him warily as he approached the bar. "What brings a man like you to these parts?" he asked, his voice trembling.
"I'm just passing through," Elric replied, his voice low and rumbling. "I need a meal and a place to rest."
The innkeeper nodded, his eyes flickering towards the door. "You best keep your wits about you," he warned. "There's trouble brewing in these parts."
Elric raised an eyebrow. "Trouble? What kind of trouble?"
The innkeeper leaned in close, his voice dropping to a whisper. "There's a group of bandits that have been raiding the nearby villages. They're ruthless, and they've already killed dozens of innocent people."
Elric's grip tightened on his sword. "I'll take care of them," he growled.
The innkeeper looked skeptical. "You're just one man," he said. "How do you plan to take on a group of bandits?"
Elric smiled grimly. "I've faced worse odds than this," he said. "And I've always come out on top."
With that, Elric left the inn and rode out towards the bandit's hideout. As he drew closer, he could see the flicker of torchlight and hear the sound of raucous laughter.
Elric drew his sword and charged towards the bandit's camp, his battle cry echoing across the plains. The bandits were caught off guard, and they fell quickly to Elric's mighty sword.
As the last of the bandits lay dying at his feet, Elric felt a sense of satisfaction. He had protected the innocent, and he had once again proved that he was the greatest warrior in the land.
With a fierce grin on his face, Elric rode off into the sunset, ready for whatever adventure lay ahead.
You can see elements that are right: the use of a fantasy-style name, the existence of swords, the attempt at dramatic tension leading to violent sword-based conflict. Nevertheless the story is terrible to the point of being ridiculous.
REH's works are in the public domain, so Chat GPT presumably can access large parts of them and map them. However, creating writing requires more than just getting a lot of the right words into the right order.
My guess is that this means we are doomed to even worse popular art than we have been subjected to already. Autotune proved attractive to the recording industry even though it is terrible, obvious, and offensive to the ear. This was because it allowed pop music stars (which now includes contemporary Nashville country music) to be selected for physical looks rather than the ability to sing. Since there are a lot more pretty faces than talented singers, that allowed the recording studios to shift more of the profit to themselves and pay less to the "talent" that is no longer talented.
AI generated visual art is already starting to pop up everywhere, and is having a similar negative function. Actual artists used to be well-paid because their talents took years to develop. Now you can generate something good-enough for free just by plugging a description into somebody's AI. It's not going to be great art, or even good art, but it will have the right elements more-or-less.
I expect we'll see a similar shift to crap like this in screenplays, especially for episodic TV but even for movies. It's free, and all they want is a consumable product to bring in money -- money they get to keep, since they don't have to pay anyone who actually knows how to produce something of genuine value. As a result, our popular entertainment will become even worse than it already is.
It's a depressing thought.
14 comments:
OK, my ego couldn't let this ride, so I issued a similar challenge to ChatGPT:
Write a private detective short story in the manner of Eric Hines
What the AI produced wasn't anywhere close to what I write. It was closer to early Ray Chandler, and utter drivel that would have embarrassed him. And apparently, the AI imitated (sort of, and badly) Chandler, never having heard of me.
Even hacks like me have nothing to fear from this iteration of AI, at least in the writing milieu.
Eric Hines
I didn't know you wrote detective stories either. I was only aware of your several works of political commentary.
Raymond Chandler's not bad company to be in.
The ChatAI is criticized for modeling language rather than reality. Though, to be fair, language models reality only moderately well, to start with, so even a perfectly congruent language model will get itself confused. But ANYhow, it seems to me the first best use of a language model is as a copy-editor of older works of language which have not yet been digitized. Say a speech being transcribe to text, to start. Anyone using "closed captioning" on TV sees how often a nonsense word vaguely similar-sounding to the actual spoken word shows up in the text. An AI might do much better at, for very least, flagging such words for human attention. Same with older books badly rendered to e-text by OCR. Way too often a substitute word or letter that doesn't work with the surrounding text is inflicted (inflEcted, for instance) upon the Kindle version. AI could and should recognize such problems.
Authors may be safe for awhile yet. But copy-editors ought to start looking for work right away.
The International Baccalaureate program is twisting itself into pretzels because of ChatGPT and the question of how to detect the use. Or should using ChatGPT be permitted? Both sides seem determined. Since I don't teach IB, I just sit back and enjoy my popcorn. There's a reason AP tests are still done by hand and proctored to a fare-thee-well. (If you use a computer because you have an accommodation, there are all sorts of rules, and the proctor watches to be certain that NO other electronics or programs are in use.)
LittleRed1
Yeah, if a grad student had turned in the Aristotle essay, I'd have given them a B+. I'd have given an undergrad an A for it; as I said, even the errors were errors I encounter in fully-trained professionals who don't happen to study Aristotle closely.
That's not bad, but it's probably close to the center mass of what Chat GPT can do well -- and since the actual point of learning Aristotle is to develop understanding, which Chat GPT doesn't do at all, it's wasted right where it's strongest. If it prevents a student from wrestling with the text and coming up with that on their own, it's actively harmful right where it's best.
The Jeep repair business was worse, but would have been a helpful tool for a beginner. I could see referring a young man who's still learning how to work on bikes or cars to it as a source of diagnostic ideas to start with. And of course the creative writing exercise was terrible as mentioned.
Raymond Chandler's not bad company to be in.
I'm working on my 6th Peter Hunt novel. Chandler would be good company to be in, were I in it. Per ChatGPT's version, however, I'm the negative of Chandler. Even without the AI's help, though, I'm not there.
Eric Hines
We were talking about this tonight at pub night. It is reminiscent of the Wordmills in The Silver Eggheads by Fritz Leiber.
https://sf-encyclopedia.com/entry/wordmills
This is all fine. By the time all the current good writers have died, almost none of the next generation will be able to read well enough to know the difference or what they're missing.
This moves us a bit farther down the road to something.
A couple of decades ago, Hollywood and the music business hired psychologists to determine what makes hit movies and music. The psychologists (despite their reputation for inanity) came through, and they produced formulas for success.
A great deal of success is simplification. Music analysts studying pop music concluded that modern music has greatly reduced range in themes, vocabulary, and rhythm and dynamic range.
No more Talking Heads or Turtles for you!
A great deal of success is simplification. Music analysts studying pop music concluded that modern music has greatly reduced range in themes, vocabulary, and rhythm and dynamic range.
It's certainly true that the most modern music genre that's actually music--'50s and '60s rock 'n roll--is quite simple in rhythms, lyrics, and range, and there's a world of difference between that and classical music. But both are terrific styles.
AI can't create either; it can only imitate. AI will actually have to achieve sentience and an emotional component to the sentience in order actually to create.
Eric Hines
"Music analysts studying pop music concluded that modern music has greatly reduced range in themes, vocabulary, and rhythm and dynamic range."
Yes, we've discussed that here on several occasions. The complexity of, say, 70s-era music is astonishing by comparison to what is being produced today.
There's also this aspect of centralized production:
https://nypost.com/2015/10/04/your-favorite-song-on-the-radio-was-probably-written-by-these-two/
The symphonic metal I often listen to seems to be retaining some of that complexity. It might be that because two of the major groups are composer-led. It might also be a way of rejecting the mainstream - "You say we gotta do chant? Oh yeah? I'm gonna polyphony the heck out of that tune, just you see!"
(Nightwish, Xandria for the curious. I also like Avantasia's more recent work, among others.)
LittleRed1
That "Elric The Barbarian" story read more like a Old West tale about a gunslinger.
I think it's important for people to realize that, in my opinion anyway, AI is just more complex web searching, with a little compilation thrown in. Like web searches, the results will only be as good as the person who receives them, and their ability to hone on what's useful and right with them, and toss out what's not. Given that most people aren't very good with web searches, I'm concerned that it will get worse with AI, as it *sounds* like it knows what it's talking about.
Post a Comment