Ab(use) of LLMs for Learning
LLMs are a great learning companion. But if you’re using them to get stuff done specifically so you don’t have to learn anything new, I have some bad news.
Avoiding the learning means treating the domain like a black box. You ask the model question after question, trying to infer just enough to get past whatever roadblock is in front of you. It’s like trying to figure out what a cube looks like from its shadow, asking someone to rotate it in various ways. At some point it’s better to just look at the cube. Touch it, even. Imperfect analogy, but I hope it makes sense.
For a lot of things - domain knowledge especially - the model works better as a learning companion than as a way to bulldoze through a task while learning nothing. There’s a crossover where it’s faster to watch a YouTube video, ask the LLM questions about what you’re learning, and then come back to the roadblock, than to spend the same hour in a mid-task back and forth trying to make the roadblock disappear.
Somewhat related: Thariq from Anthropic shared an eli5 skill that people there have been using a lot. He uses it for things like /eli5 how does this module work or /eli5 what caused this incident. To install:
claude plugin marketplace add anthropics/claude-plugins-communityclaude plugin install eli5@claude-communityThen /eli5 whatever you’re stuck on.