ChatGPT Is a Blurry JPEG of the Web
Ted Chiang's New Yorker essay arguing that a large language model is a lossy text-compression of the web: it stores statistical regularities and reconstructs missing text by interpolation. Its hallucinations are compression artifacts — plausible enough that catching them means comparing against the originals.
- Author
- Ted Chiang
- Published
- 2023
- Original
- Read it at newyorker.com