I saw some morecantile/mercantile in the article, so I will plug a side project of mine “utiles” for anyone needing tile util(e)s: https://github.com/jessekrubin/utiles
I'm one of those people who is mostly not-funny, but is occasionally quite funny. I think that if I were given timely material, I could execute it well. I think this would work best with one-liners, where there's less of a need to get the timing right between sentences.
For example, if you'd been having a conversation about Topic A, then later Topic B, and there was a joke involving a double meaning of a word that relates to both A and B. Wrap that up in a one-liner, put it on a Google Glass, and a decent number of people could deliver it well enough to get a laugh.
One of the things that surprises me about our hopes for AI is the belief that it will somehow imbue us with creative powers that we don’t have.
The idea that GPT is
going to make you funny is a bit like thinking you could be John Lennon if GPT would teach you how to play guitar.
GPT might help you pass off some bit of wit in an email, but any art it creates (comedy is art) is going to be circumscribed by its inability to be truly creative rather than merely copy/regurgitate existing material. And the reason art is hard is because people are extremely perceptive about what is actually creative/funny/surprising. You can’t fake it.
I dunno, it seems like it would be good at finding linkages between different topics and then coming up with a word or concept that relates to both. This is at the heart of much humor.
Will it have a good hit rate? Surely not at first, but of course it will get better over time. And just like with graphic arts, much of the value is that it can create lots of options, and you can choose to just use the ones you like best. It's like a wacky brainstormer that is sometimes right-on (or inspires you to create something that you never would have thought of).
That sounds a bit silly. Why can't funny be "faked?" Plenty of successful comedians practiced and had to learn to be funny. So there's clearly a method to it.
Sense of humour (intelligence) is about the ability to find improbable, but insightful things. I believe spitting out patterns extracted from training data is not capable of producing anything close to that.
by your definition, aren't you just asking it to find the patterns that're uncommon then? given the amount of ram it has access to, that should be at least theoretically possible to generate uncommon patterns and produce something approximating humor.