Your son is following much more interesting classes than I was at that age!
disclaimer: it has been quite some time since I was 13 so I can't really gauge how difficult this is! Theory-wise, you can't really do better than 3blue1brown (but same disclaimer as above, might require some tricky maths): www.youtube.com/watch?v=airc...
YouTube video by 3Blue1Brown
For any modern type of encryption it should be nigh impossible to determine the encryption method from the ciphertext (by design). Classification of e.g., images might be a lot simpler and pretty cool? Keras makes it quite easy in combination with something like www.youtube.com/watch?v=BfCP...
YouTube video by Andreas Zinonos
Me, every time @cjungerius.bsky.social suggests Bayesian Models.
Excited to share my primer on simulation-based power analysis using R, Python, and Julia! Based on @debruine.bsky.social@kirsten-adam.bsky.socialcjungerius.github.io/powersim/
The second preprint of my PhD with @haslagter.bsky.socialosf.io/exgvu
While I have some minor gripes with this - I feel like MCMC has won the Bayesian inference method war, and the focus on belief updates ignores the value of posterior predictive checks (*how* do models fail, and why?) - in the abstract I really like this take on adverserial collaboration.