"A Woman of Mystery" by GPT-3 and Adam Carlton

GPT-3 is the text-generation system recently developed by OpenAI using "175 billion parameters, two orders of magnitude larger than the 1.5 billion parameters in the full version of GPT-2".

The easiest way to try it out is to use the “custom” option (6) of AI Dungeon with the Dragon engine enabled. Dragon is GPT-3, the default Griffin is the less-powerful GPT-2.

You can seed the AI engine with a starting sequence and then it will generate continuations. Here is a story Adam Carlton developed with GPT-3 using the AI Dungeon app and the seven-day free subscription for Dragon, published on Booksie.

One author is in bold, the other is in plain text. Which is GPT-3?

---

A Woman of Mystery

My wife died recently. I am sad and lonely. I sit in a bar drinking and reading the paper. I see an attractive girl about my age in a short cotton dress. She holds my gaze. I invite her over and we begin to talk. I have a feeling this meeting was intended though. What does she want?

"I'm sorry, but I don't know anything about you."

"What do you mean?"

She says that she is from a small town in Texas and has never been out of it. She also doesn't seem to be very well off. "Well, I've got money for gas and food so I can get back home tomorrow night." She then goes on to say that she's going to go to the movies with some friends.

But she then contradicts herself and asks whether I might have a better proposal. I think furiously, keen to spend more time with her.

"You're not really interested are you?" I ask.

"No, no I guess not." She replies. "But if you were, would you still come here?"

After spending some time talking with her, we both decide to leave together. We walk along side each other and then suddenly she turns around and looks at me again. "Would you like to go back to your place now?" she asks as if she just made up something completely unexpected.

We go back to my place and I sit her down on the couch. "Do you want something to drink?" I ask her.

"Some juice or soda?"

She smiles. "I'd love a diet coke, thanks" she replies.

I go into the kitchen and come back with two diet cokes. She takes one and we both sit down on the couch. It feels strange to me that she is here.

She starts to ask me about myself. The conversation rapidly moves to my financial situation. Just as I start thinking she's a gold-digger she tells me she's feeling warm, "Do you mind if I take my dress off?"

"Oh, no. Please do."

She stands up and slips the dress over her head revealing a lacy black bra and panties. She sits down and takes a drink of diet coke before saying, "If you really want to help me out, I could use some money for food and gas."

She looks at me with her big green eyes and hopeful face. I'm really starting to like this girl.

"I don't think I can afford it," I say.

She comes across to me with her big eyes and slowly kisses me. "Are you sure?" she says.

She kisses me again and rubs her body against me. "Oh, you're just so sweet," she says.

We have wild passionate love making. Afterwards we're laying in bed when she says, "I guess I should go home now."

I tell her she can stay the night and we fall asleep like that. In the morning I wake up before her and have to go to work.

When I return the house has been ransacked, but she has left a note.

"I'm so sorry for stealing your money and jewelry. I have to get back home as soon as possible.

Thank you for the most wonderful night of my life!"

--- END ---

Adam observes: 'I don’t think serious authors will be quaking in their boots quite yet. But if this was submitted as an assignment in a middling college by a mediocre student it would get some kind of mark, almost certainly better than the unaided efforts of some students.'

---

In my view this is a kind of IQ 90 story. It makes sense in its muddling-along way, but lacks dynamism, narrative drive and any psychological validity whatsoever. However, experience has shown that that's not necessarily an obstacle for commercial or academic success.

The deep problem is just obvious. As I observed in this January 2020 post, "It's just a better parasite...", these sophisticated contextual text-shufflers are leveraging contextualised human literary productions without any internal model of the natural or social world, or human agency. Don't look for personalities here!

I know that quantity has a quality all of its own, but in terms of story development (plot, setting, character) Adam tells me that working with GPT-3 is an exercise in futility: you can never get going - you are continually derailed.

But for many applications where text has to be produced: who cares?

---