2025 - 10 October
Will the explainer post go extinct?
Q. Are “explainer” posts done for? If you want to re-learn about the chain rule in calculus, should you spent any time on their popular machine learning blog that wrote 12 pages on it?
The article says that once AI is “accurate” enough, all other reason why you might read one fall away and the explainer genre is toast.
I agree that the answer should be “no”, except in the few cases where the author has produced something very high quality, with feedback from peolpe who have tried to learn from it. Really the closest genre of internet writing that I think will survive is the “eplain what I did” or the “follow my journey” posts. A prime example: [This detailed blog on created a homemade SAR autofocus algorithm] on his homemade drone. It’s extremely impressive. He has a decent amount of explanation in ther, and many of his blogs have accompanying code on Github. But it is not a post where you could walktz in with a little background knowledge an expect to build one yourself. In fact, it’s more of a target for someone like me, who has worked in a NASA SAR algorithm group, to be able to replicated it. There’s a lot of signposts, but it would be a backcountry learning expedition, not a hand-held walk in the park.
But that’s part of what makes it interesting — it’s such a complex project that it would be very dull to read an explanation of every single step, and in fact it would turn into a semester-length course to lay it all out. That work would never happen, so this version that exists is much better.
My current thinking on the useful genres related to the explainer:
- Finding the highest quality traditional teaching material (e.g. https://ocw.mit.edu/) is still very valuable. The good courses on there have been planned and refined for decades, targeted at a specific audience, but created specifically as teaching materials.
- The missing part of taking that course is the feedback from professors and TAs- and this is not where using LLMs is key. There is basically no online course where you could stump a LLM with a homework question. This means that all of those great physics courses which stubbornly kept off their answer kets are nearly as good as the problems with solutions! In fact, it’s likely that the LLM solution may be better, since you can ask it to be as verbose as you need. Making high quality solution keys for full courses is incredibly time consuming — it’s why courses without public answers are understandable, even if they are annoying.
A reminder to
- find the missing channels of information when you find yourself getting second hand stories
- cut out low SNR channels (and meetings!)
- urgent is not important