Lemons
I made this suggestion, publically, a few places now, and have been using the term daily to refer to them.
No longer should you refer to Large Language Models as “LLMs.” Instead, call them “Lemons.” It‘s easier to say, and I challenge you to not have fun while saying it in this context.
There are parallels to LLMs which make this more than just a silly wordplay. For instance, lemons, the fruit, really can’t stand on their own. You can‘t really eat them, and it sucks to get lemon juice in your eyes or in a cut (although, it is a natural antiseptic; but painful).
LLMs have to be prompted. They’re an ingredient to a broader goal. They might be able to “flavor” your work in impressive ways, but it‘s likely disasterous if you try to squeeze too much into the pot.
Going back to the silly (maybe “troll” is a better descriptor) aspects for a second, “Lemons” works great if you’re in the “Never LLMs” club.1 “Those things don‘t work! They’re lemons!”
Footnotes
- I‘m against LLMs for lots of reasons, but ignoring them completely is unlikely to result in the best career outcomes. I hope we start to redirect Large efforts to Smaller language models targetted at solving specific tasks well. back