<
https://pivot-to-ai.com/2024/07/14/ai-does-stand-up-cruise-ship-comedy-material-from-the-1950s-but-a-bit-less-racist/>
"Google DeepMind went to Edinburgh Fringe in August 2023 to see how twenty
working stand-up comedians could use large language models. It didn’t work out
so well:
most participants felt the LLMs did not succeed as a creativity support
tool, by producing bland and biased comedy tropes, akin to “cruise ship
comedy material from the 1950s, but a bit less racist.”
It turns out that LLMs heavily tuned to be inoffensive aren’t very good at
telling jokes that use satire and offense to punch up, not down. The paper from
DeepMind lists this as a fundamental limitation of current LLMs.
These comedians were already writing with LLMs — but didn’t want their names
revealed: “I have friends who don’t talk to me anymore, because they learned
that I have an AI show going on here” or “I have a friend who is deeply upset
that I was using generative AI in my flyers.” Maybe take the hint?
The stand-ups wrung their hands over the ethical concerns — plagiarism, lack of
diversity in the training data, and whether it was right to just clone
someone’s style.
But they wanted to keep using the LLMs, so they decided: “if we didn’t put all
that stuff in there, it wouldn’t work as well.” Never mind that it did not, in
fact, work well.
The paper concludes: “These are thorny open questions, left to the readers to
address.” Yeah, thanks, guys."
Cheers,
*** Xanni ***
--
mailto:xanni@xanadu.net Andrew Pam
http://xanadu.com.au/ Chief Scientist, Xanadu
https://glasswings.com.au/ Partner, Glass Wings
https://sericyb.com.au/ Manager, Serious Cybernetics
