This list is reproduced from the latest issue of my newsletter, Time’s Corner.
You may have seen articles about the copyrighted text and images that AI models spit out. I believed AI companies were guilty of massive copyright theft, until this article by Cory Doctorow calmed my outrage somewhat. He explains that Midjourney (an image generation site) keeps only about one byte of info from any given image it sources: “If we’re talking about a typical low-resolution web image of say, 300kb, that would be one three-hundred-thousandth (0.0000033%) of the original image.”
Why AI will never topple the film industry: “To put it as plainly as possible, every single time that Shy Kids wanted to generate a shot — even a 3-second-long one — they would give Sora a text prompt, and wait for at least ten minutes to find out if it was right, regularly accepting footage that was subprime or inaccurate…”
Mary Harrington explores the scary idea that, thanks to artificial intelligence, using social media may become a form of social activism.
Robin Sloan asks whether large language models are in hell: “The model’s entire world is an evenly-spaced stream of tokens — a relentless ticker tape. Out here in the real world, the tape often stops; a human operator considers their next request; but the language model doesn’t experience that pause. For the language model, time is language, and language is time. This, for me, is the most hellish and horrifying realization.”
Related to the above, James Bridle asked ChatGPT to design him a chair and, unsurprisingly, discovered that AI has no idea what an actual chair should be like.
And lastly, Samuel Arbesman suggests we explore the “story world” of AI, since so much of the language input is in story form.
Christian Leithart 