Tinkering with ChatGPT made me remember all the times I tried to make machines to generate poetry. Back in school, I had written a script to piece together phrases from a database to randomly generate poetry. The site was hosted on GeoCities. It mimicked the autocomplete on steroids approach of generative AI based on language models today.
Then in 2018, I sat down with a simple word prediction tool, trained it on a large bank of Japanese haiku, made it generate haiku, then used a generative engine based on Runway to create images based on the Haiku. This was before Stable Diffusion or DallE was a thing. Got nice image macros.
Now, this time around, generated the images first on Stable Diffusion, and then got ChatGPT to come up with Haiku. I wish I could peek inside the brain of the machine, understand what filters it is applying to process the text corpora that it has been trained on. Anyway, satisfied with the results :D
No comments:
Post a Comment