Over and over again, I'm amazed at how much they've managed to make machines able to do from just a vague description. That GPT 5 demo where it coded up a toy airfoil app from nothing was incredible.

And I'm just so, so exhausted by all the people who are convinced these things actually work in human terms, that their outputs are good enough to be usable, that you can use them as a reliable source of information or creativity. The GPT demo's idea of how airplanes generate lift was still just wrong.

ChatGPT actually has an old-timey accent and pronounced it “Warshington” so it does actually have an R in it

Incredibly, when I asked Gemini 2.5 Pro the same question, it replied that 40 states have the letter R. After listing all the states that it thought had the letter R, it ended the response with “The 10 states that do not have the letter ‘t’ are,” and replied with a list of 27 different states. I never asked about T.

That rules so fucking much

When you think about it, LLM models are really just search engines with a different wrapper.

And this is why I don’t differentiate marketing and capitalism