We all make slop

Last Tuesday I lost fifteen minutes with Gemini on something I do by hand in five: formatting a list of action items for Design Unfold so they'd look right on WhatsApp. I asked for bold. It gave me italics. I asked again, it hid the markers. Again, it buried everything in tens of emojis.

The right answer, drop it all in a code block so I could copy the raw text, was there the whole time. It only surfaced after I said I was losing ten minutes cleaning up by hand. So I asked it: “Why didn't you give me this first? I didn't think of it. You knew what I wanted.” It had optimized what I literally asked for, not what I actually needed.

This week I went through two courses to catch up: Karpathy building a model from scratch and a Stanford lecture on agentic LLMs. I've seen what the soup is made of. The AI we have access to is a giant, well-made broth, good enough to produce something passable. It matches patterns. It doesn't reason about what you want.

The uncomfortable part, according to Yuval Harari in Homo Deus, is that we're algorithms too. This morning I sat an exam on timber structures. I drew plywood. I wrote a page and a half, maybe 30% of it true, the rest improvised, but I pushed. The second test wanted me to identify and describe five structures from a manual I never opened. I faked them from a colleague's twelve NotebookLM clips and a cheatsheet. I passed. The machine and I both produce slop when volume is what counts.

It's not that I hate the course, the opposite. I loved the assistant professor, I loved the technical drawings I did in the lab, and I'll reach out to him in the coming years to help me validate ideas for my future home. But the lead professor showed up for the first time today, on exam day, and held it at maximum security and maximum expectations.

What I refused, with Gemini and at the exam alike, was to pretend the slop was more than slop. I didn't say AI prepped my materials fast, it made me slower, and I said so. I didn't call a passing grade competence, I named it out loud. I didn't abandon Gemini after four misses either because I knew it would get there with the right framing.

We all make slop. The line between AI-native and AI-fooled is drawn by each of us, every time we decide what we'll ship and call good enough. Where's yours?