How Did Thirty Minutes With a Language Model on a Fresh Rails Project End Without the Test Ever Running and With a Ton of Software Installed?

On the show, a weekend try to get a language model to run the test on a fresh Rails project failed. Thirty minutes later the test still had not run, and the model had installed a ton of software.

The speaker had been pretty bearish on generative AI for writing software, then on a day off opened an editor to see whether the model could even run a new Rails project. The instruction was to run the Rails project and run the test. Thirty minutes later the test still had not run. The model had installed a ton of software and used up a bunch of space because it could not figure out what was wrong. A lot of old software for older versions of Ruby and older versions of Rails drove the model crazy, so the session ended with the words, you and me, buddy, it's been fun, but we're done for today.