#MeWriting  LLMs are good for teaching some skills like writing, coding, and drawing by providing infinite examples.

No, I do not want to live in a world where all content and code is written by LLMs. Or even worse, processed, edited, and approved by LLMs. Even if they were perfect oracles, writers, and coders, I don’t want that. I trust the process of humans creating great content, even the best content.

I also don’t want to live in a world where most of the content and code is written by crappy or lazy humans.

The struggle has to be real AND human to earn my trust and approval.

However… people have to learn to write and to code. LLMs are better at these things than a sizable chunk of people. Even zero struggle LLM output is better than the high struggle output of a good chunk of people, perhaps a majority, maybe even 90% for some niches.

I also don’t want to settle for a 90% world. This technology, which finds average language use, will not achieve above some threshold without having lived experience that isn’t written down and in the training corpus or (better) context window. It can’t, by definition, consistently reflect the style or taste or wisdom of an individual.

Part of those “10,000 reps” is getting to 90%. You can do that by mimicking what LLMs can generate, using those not great outputs to level up while you’re even less great.

I’m not talking about putting LLMs in charge of the learning process, just generating the reps. LLMs suck at orchestration. I hope we can throw most of those reps away as the learner gains competence and confidence. So eventually, that person can create great writing, code, etc. on their own.

If this interests you, I’d love to have a conversation with you. Drop me an email.

Since I still suck at drawing, I asked Grok to sketch me a face that might be a reaction from everyone else to my heresy.