I want to document where LLMs are at, today, partly because I don't know of any single place that brings all these examples together. I thin
Four years ago we were laughing about how AI couldn't draw a human with the right number of fingers. The state of the art can generate accurate QR codes. The most likely outcome is that LLMs continue to improve; there are numerous improvements already being developed, and no sign of things slowing down.
An attempt to document/approximate where LLM progress is at right now (July 2026). It's art can pass the Turing Test - people can't reliable distinguish non-slop human vs AI output, even if a lot of AI output is in a more identifiable slop style. They've solved multiple open problems in math. They're outperforming programmers in contests, and they're also somehow starting to beat out writers too.
Even if you oppose LLMs, it's useful to be educated on how far they've already come, and I expect we'll see equally significant progress over the next 3-4 years.
