Last month I wrote about the new acronym AI;DR (AI didn’t read). The idea was that people were tired of having their email queries answered with AI slop that even the responder probably didn’t read. As I said in that piece, sending someone unedited, possibly unread, AI slop is profoundly disrespectful and will cause you to lose the goodwill of your victim.
Of course, that’s just my opinion but Bryan Cantrill, the CTO of Oxide Computer Company, has a well thought out post concerning the reader revolt from LLM generated writing. He notes two important points:
- Readers can tell AI writing from the real thing.
- Readers do care.
Then he points to a blog poll by Cynthia Dunlop that showed that 78% of developers stopped reading as soon as they detected AI generated content and that an astounding 71% wouldn’t read anything else from the same writer. Talk about sacrificing the reader’s goodwill. I urge you to read that post and the previous poll that discusses why people use LLMs to write. They’re both worth reading.
His next point is to recall how email was saved from the spammers by improved filtering. Once spam filters got really good at identifying spam and removing it, sending spam became an uneconomical endeavor. Cantrill thought that since it was so easy to identify AI generated writing, it should be possible to automate the process as it was for spam filtering. That turned out to be harder than he thought but he has found a solution. You can check his post for the details but the end of his story is that now all public Oxide Computer Company writing must pass through this filter and come out clean.
Perhaps the self imposed exile of those using AI to write their content and the continued improvement of filters to automatically identify it will slow down its use just as happened with spam.