snarxiv.org was launched in March 2010: a site which generates random imitations of arxiv abstracts.
This blog was launched in June 2011 with an experiment: What happens if you take snarxiv abstracts seriously? What might the paper accompanying a given abstract, actually be about?
In June 2020, OpenAI began to make GPT-3 available to the world: a "language model" trained on Internet text, which could write a whole essay, or other verbal production, given a short "prompt" to set it off.
Now in June 2022, I have used GPT-J, another model inspired by GPT-3, to write a whole fictitious physics paper, using a snarxiv abstract as the prompt. Here is the result.
The individual sentences make sense, but the resulting paper is not coherent on scales greater than a paragraph. But somehow I suspect that in less than ten years, you'll be able to input a snarxiv abstract into an AI, and almost instantly get back an essay which really does its best to deliver coherently on the promised content.
No comments:
Post a Comment