In addition to the other great suggestions, point him to Karpathy's YouTube channel[1]. Karpathy has an approachable communication style.

Here's his "1 hour intro to LLMs" video: https://www.youtube.com/watch?v=zjkBMFhNj_g

1. https://www.youtube.com/c/AndrejKarpathy

Not an expert, but maybe using RAG/embeddings on the on-disk wikipedia would be better than finetuning on wikipedia?

Most decent LLMs probably were already trained on wikipedia, that doesn't stop them from hallucinating when asked questions about it.

Use a model already trained on Wiki[edia using llamafile.

You can download llamafile and several models, put them on a USB drive or hard drive, them send the drive to him via DHL.

Would it be possible to ship him a Starlink terminal? Internet access could do wonders for a young interested guy like that... And he could share that connectivity with people around him too.

What kind of GPUs does he have?