I’m undertaking a 1000-day reinvention project, blogging here daily to track my progress. In Wednesday Writing, I consider my writing practice and skills and how to improve upon them.
On Twitter, freelance editor Tricia Dearborn writes of the problems that arise she sees in manuscripts that have been structured, organized, or refined using gen AI:
Turned out these manuscripts had structural problems. But that wasn’t the issue. Manu manuscripts have structural problems.
The issue was that there was no way to fix the structural problems.
When a human author has written a work, when there’s a unifying intelligence behind it, it’s easy for an experienced editor to work out what the problem is, get a sense of what the author is trying to do, and (often) see how it could be fixed.
When a work has been generated by AI, there is no unifying intelligence.
Elsewhere on Twitter, I’ve seen critiques of generative AI and reports of problems related to its use, going beyond the typical “it just generates slop”:
- LLM-based agents are failing for general-purpose problem solving because they don’t select actions to maximize the user’s expected utility but rather are optimized to generate the next token in a string of text
- Source code control platform GitHub is “hitting a breaking point as AI coding agents flood the platform with far more commits, pull requests, searches, and CI jobs than its older infrastructure was built to handle.”
- OpenAI’s coding agent Codex based on GPT-5.5 has an instruction telling it “Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creates unless it is absolutely and unambiguously relevant to the user’s query.”
We may be entering the trough of disillusionment in the generative AI hype cycle.
But maybe not, if you look at stock returns for AI-related equities vs the rest of the U.S. large cap market:
I’ve stopped subscribing to any premium version of Claude, ChatGPT, or Gemini. I’ve found their performance lacking. I rarely use them to help me write, though I do use them to learn about topics I don’t know much about.