![]() ![]() It still requires a human to feed, steer, groom, and clean up after it, deduplicate repetitions, filter stupidities, fix misspellings, grammatical mistakes, glitches, contradictions, then rationalize and weave it into a coherent story. So far it’s only cost about five or ten dollars in GPT-3 tokens (they charge by the word), but it’s generated a LOT of value and ideas and laughs! It might be useful as a tool for generating content for a web site (like “Silly Strains” or “Funny Shit Stoned People Say”).īut it probably should’t used literally and unedited, however used more as inspiration and exploration, like Brian Eno’s Oblique Strategies, or Roger von Oech’s Creative Whack Pack. I’ve also been experimenting with having GPT-3 write eloquent flowery strain descriptions, and classify strain names into categories, with some success. Here is the start and end of a long list of cannabis strain names it produced, most of which are actually real (which I discovered by googling), and many of which I’ve never heard of before! The 435 Names of God It turns out that if you give it an alphabetical list of names, it’s really great at playing the “This Cannabis Strain DOES Exist” game, and exhaustively enumerating the correct names of real cannabis strains!Ĭan anybody explain why GPT-3 is so good at alphabetizing? That seems to set it on a linear trajectory that densely enumerates the possibility space much more deeply than randomly sampling, which tends to get distracted and spiral out of control. ABCDEFGHIJKLMNOPQRSTUVWXYZBGLEBULOBZOINK! Using generative adversarial networks (GAN), we can learn how to create realistic-looking fake versions of almost anything, as shown by this collection of sites that have sprung up in the past month. ![]()
0 Comments
Leave a Reply. |