Slop & Anti-Slop

May 18, 2026 ·

There's been lots of use of the word slop lately, describing large outputs of AI-generated things. Text, media, code, what have you. Within code, but also generally, the claim is that so-called slop is entirely unusable and pointless, only worth hype and lacking substance. I don't think that's fully correct and more so a defense mechanism or some kind of rejection of the new technology. I've seen this game so many times now that I just know most people reject innovation and technology while coming up with increasingly sophisticated reasons for doing so, most often as some kind of taste and intellect thing; sometimes arguing for morality and ethics or aesthetics. Any kind of argument to make one sound better and more elite type thing. I think Vitalik is a highly balanced thinker and I liked what he said in his latest blog post, that LLMs are increasingly also good at bug finding and fixing. So the cure for slop is...anti-slop mechanisms also generated by LLMs. Neat.

In other news, I came up with a first pass for the book project I mentioned yesterday. It's another example of the kind of things possible with LLMS, that I made an in-house Google analytics the other day, and generated a functional first-pass for a book idea in half a day today. I am not happy with it yet, but it's indeed an impressive first-pass. As-is, it would make for slop. But to Vitalik's point, I can make LLMs hunt down bugs and de-slop the thing with more effort. It's both true that I am not bothering to look deep into the code, which is what bothers critics, but also that I am able to produce things unthinkable before. The claim that these productions are slop or only surface level creations are addressable and in fact next problems to chase. In any case, in the spirit of openly building I wanted to share first copies I just created.

cover.pdf

interior.pdf