1. You can use LLMs to help you learn faster.
2. Review other coders' code to help you learn.
I fully agree with the second one.
I only cautiously agree with the first one, and I'm pretty skeptical of the formula they give in the article. Traditionally, education is accomplished by studying primary and secondary sources, and then by practicing. LLMs don't somehow hack around that core idea, but I think there are ways they can streamline the process. What I propose is that you ask pointed questions to the LLM and have it provide sources in its answer, which you then go and check and read parts of. You still need to evaluate whether each source is trustworthy, but the LLM could save you the initial trouble of tracking down the sources in the first place.
My general advice here though is to tread very cautiously. I think there's a real danger of fooling yourself into thinking you did all of this learning when in reality you're outsourcing a crucial part of the learning process (such as critical thinking) to the LLM, and the really problematic part is that the LLM is pretty much the perfect tool to trap you in that way.
But when you’re new on a topic, it can take hours to nail the solution by yourself. I’m thinking of science & math practice problems. If you skip ahead to look at the answer, you can use it to work backwards, to understand where you were struggling and going wrong.
Crucially, you’re not just copy/pasting the answer. But taking this “shortcut” early can definitely be helpful. It’s very easy to spend a lot of time going down the wrong path. If you can take the shortcut when you sense that’s happening, and really try to understand why that wrong path isn’t right, I think the shortcut can help speed you up.
I think the same applies with LLMs. It’s important to not just have the LLM do it all from the start. But when you start getting stuck, having it give you recommendations is a good way to get tutored, so long as you’re really thinking about it.
downrightmike•5mo ago