Perhaps I should consolodate all of my previous posts linking to articles discussing various aspects of why LLMs are so unconscionable to me:
- Open Letter against LLMs in Education
- Old Man Yells at Cloud
- "AI is Destroying Learning Itself"
- It's rude to show AI output to people.
This post is adding to their number. William J. Bowman recently wrote up this excellent article which really delves into whether LLMs could actually be considered useful. A great point he made is that as the models get "better", they actually require more work to be useful, as finding the problems in their output becomes harder and harder.
I was also relieved to see that AI is even less popular than ICE in the US (although I was slightly annoyed by the headline of the article, which claims that people hate AI more than ICE, a claim not supported by the poll, which does not measure intensity of dislike).
Also Anthony Moser wrote this excellent piece that was very cathartic for me to read.
Finally, I was glad to see this list of software that has pledged not to accept AI contributions and not to integrate generative AI features. While most of the software on this list is not useful to me at present, I hope it takes off and grows. I do have a very brutal approach: if any software I use tries to integrate AI features, I immediately cut it out of my workflow, no matter how essential. And, if a project accepts AI-authored contributions, I'm on high alert for any possible replacement. This is, for example, why I stopped using ghostty.