This site uses generative AI to create fortunes.

You know, like the kind you get in fortune cookies.

Click that random fortune → link and get a new fortune. Click around till you find one that you enjoy. Share it with friends, if you like.

That's about it.

For fun!

Years ago — around the time Markov bots were popular — I experimented with computer-generated fortunes. I built a few prototypes and shelved them.

More recently, my LLM explorations jogged my memory. I figured: "Why not treat genAI as a source of randomness?" Instead of getting a computer to generate pseudorandom numbers, I'd let a machine generate pseudorandom text.

I invoked various LLMs to create the fortunes. If the LLM providers employed people behind the scenes, furiously typing out text in response to the prompts, that's on them.

(Before you laugh: the "AI is really just people behind the scenes" has allegedly happened before. So who knows?)

But in all seriousness: I wrote a bunch of prompts. What you see here is what the genAI models sent back.

Nope.

I have removed some of the generated fortunes, however. Because, yes, I performed a light review of what the LLMs sent back.

(If you've seen any of my other work, you may recognize "Check the Bot's Output" as my theme.)

Fortunes subject to removal include:

• Potentially objectionable content. I haven't seen any of this just yet, but I check just in case.
• Places where the model emitted a stream of gibberish control characters. You'd be surprised what sends the LLMs into a tailspin...
• Cases in which a model refused to generate a fortune because the prompt tripped some internal filter. These were fairly innocuous prompts, mind you. Some providers' filters are a little sensitive.

The fortunes that made it through that filter are provided unaltered. And that includes some fortunes that were clearly glitches committed by the LLM. Click around long enough and you'll see what I mean.

Yes.

The fortunes came from the bots, but the bots based those fortunes on the prompts I issued.

(Some of those prompts were inspired by conversations with friends. A very special thanks to N, Thot Leader, and a few people who'd prefer to remain anonymous.)

Hi! I'm Q McCallum. My main website has a more detailed description, but here's a short version:

I work as an independent consultant, researcher, and writer. I've been working in (The Field We Currently Call) AI since the early days of predictive analytics and data science.

I occasionally share thoughts on my blog, Bluesky, and O'Reilly Radar. I also publish a newsletter called Complex Machinery about the intersection of risk and AI.

I find randomness to be a pretty interesting concept.

Not at all.

GenAI is good at some things. It is terrible at other things. The key to genAI, as with any other tool, is to use it where it's a good fit.

Thus far, it's demonstrated some ability to write short strings of text that I can pass off as fortunes.

Yes! You can follow Fortune Ex Machina on Bluesky.