A Note on LLMs
Someone asked on Mastodon if I was using AI to write this series of posts. My first response was, “Ouch.” My second was to say that yes, I’m using Claude in a couple of ways:
-
To check for redundancy and overlap. I have a tendency to tell the same stores repeatedly (and yes, that is my wife and daughter you hear sniggering in the background). LLMs seem to be pretty good at spotting this.
-
As a better search engine. I know a lot less than I want to about history and politics outside Europe and North America; prompting Claude to give three examples of cooperatives from other parts of the world is a lot more productive than searching on Google (in part because Google’s results have been getting steadily worse).
I think (at least, I hope) the question was prompted by the volume of posts rather than by my writing style. I am a little peeved that em-dashes and semi-colons are now taken as signs that text was generated by an LLM—I’ve been using them since high school, and would hate to abandon them because they make people think I’m a robot.
Stepping back, I first had the idea for these posts in 2017. In the wake of Trump’s inauguration, I realized that for every Software Carpentry workshop that taught coding to researchers, there should have been a workshop that taught politics, economics, and sociology to programmers so that they would understand how the world actually works. I tried to crowdsource the necessary material, but that didn’t work out.
Nine years and a lot of reading later, unemployment is giving me time to collect my thoughts, I think I finally know enough to write something that is less wrong than it would have been. I still don’t think I’ll ever try to turn these posts into a book, but I hope that they’ll prompt people who actually know about this stuff to take the time to correct me.