Training a trillion parameter model to be funny

I make a project for evals and fine-tuning and our default example task is a joke generator. It's a fun demo, but more importantly it's a really good use case to show how evaluating and optimizing LLMs is hard.

- There are a dozen plus common failure modes. How you split setup/punchline. Tropes. Toxicity. Template reuse. Each one needs a good eval.

- Datasets are hard: there's not much off the shelf, and as this author points out scraping gets a weird mix of quality.

- Models are really bad out of the box at humour.

At the end of the day it's just a hard problem that takes a lot of work and still isn't solved. GEPA prompts help, if you have good evals. Supervised fine-tuning works a little bit, but only if you training on a chain-of-thought thinking phase. We have a new evaluation builder that uses examples of edge cases for alignment, and jokes require the most iteration and feedback for refinement.

If you want to try it: https://github.com/kiln-ai/kiln

Is writing in all lowercase funnier?

...this is actually a really interesting thought.

The act of writing in lowercase is not, in itself, funnier. But writing in the training set that is in all lowercase is _probably_ going to be the funnier writing.

Considering modern pundits online, "lowercase" is usually the case of the humourist. Lowercase also tends to be the case of sarcasm, almost exclusively deployed to be funny.

So it would make sense that models attempting to select for funny would also write in lowercase.

I mistakenly read this as training a trillion parameter model would be funny...at least I chuckled

Is the comedy that these jokes suck?