[whacked_new]

Circa GPT-3.5 to GPT-4o I was involved in some research in figuring out how to make LLMs funny. We tried a bunch of different things, from giving it rules on homonym jokes [1], double-entendre jokes, fine tuning on comedian transcripts, to fine tuning on publicly rated joke boards.

We could not make it funny. Also interesting was that when CoT research was getting a lot of attention, we tried a joke version of CoT, asking GPT4 to explain why a joke was funny in order to produce training set data. Most of the explanations were completely off base.

After this work, I became a lot less worried about the GAI-taking-over narrative.

Funny is very, very hard.

[1] without a dictionary, which at first seems inefficient, but this work demonstrated that GPT could perfectly reconstruct the dictionary anyway

[astrange]

The GPT3 base model was pretty funny if you like nonsense. Instruct tuning and RLHF seem to destroy it when they recalibrate everything.

[lofaszvanitt]

There are very good, less known, models that produce funny and highly creative outputs when nudged in a good way. Premier models are just plain meh in this space.