I am trying to understand LLMs conceptually well enough to be able to predict their capabilities (and limitations) when it comes to generating code. Is that even a sensible goal? Are there good resources?

So far I've looked at:

1. Vibe Coding, by Steve Yegge and Gene Kim (https://www.amazon.in/Vibe-Coding-Building-Production-grade-Software/dp/1966280025). This has some practical examples and many guidelines. But there is not much theory and this does not explain LLMs conceptually AFAICT.

2. Build an LLM from Scratch, by Sebastian Raschka (https://www.manning.com/books/build-a-large-language-model-from-scratch). Seems in-depth. But I don't really want to build an LLM.

3. AI Engineering, by Chip Huyen (https://www.amazon.in/AI-Engineering-Building-Applications-Foundation/dp/1098166302). This seems promising, although it is not coding focussed.

Perhaps something like How Claude Code Works (https://code.claude.com/docs/en/how-claude-code-works) but fleshed out in more detail.

Thanks.