I gave two conference talks this summer, and both recordings are online now. They come from very different corners of the work I do, and I’m honestly pretty proud of that: one is about type systems, the other about building tiny machine learning models by hand. Both go fairly deep, and both have live coding in them, which is the part I enjoy most in all of my talks, and the part that people seem to get sweaty hands about. They were both long (30 and 45 minutes, respectively) and within a week from each other, which is the part that I was nervous about.
Both went fairly well, if not exactly perfect. The failure modes are not anything technical, but mostly stem from the fact that I always want more time to go deeper, and sometimes compress things unduly when talking through things live on stage. Still, they hold up I think.
The first one is from the WeAreDevelopers World Congress on July 9th, and it is called “Type Systems You Might Not Know (But Will Love)”. It’s an argument against reducing types to the static-versus-dynamic question, and a tour through a handful of type system features that rule out whole classes of bugs by making opinionated choices about how the world works.
One warning before you click: the audio on that recording isn’t great. It’s understandable, but you’ll probably want headphones, and I’m sorry about that. WeAreDevelopers is more of a expo than a conference, and I gave the talk in one of the main expo halls, with the clip-on too close to my mouth, so both clicks and a background hum are audible.
The second one is from EuroPython on July 17th, and it is called “AI Architecture Katas: Learning by Building Small Models in Plain Python”. The idea is that you can learn a lot about modern LLM and AI architectures by building small, runnable versions of them with nothing but Python and NumPy. We build a little Transformer end to end, and then go looking at multi-scale modelling and state spaces, always staying small enough to keep the whole thing in your head.
If either of them is interesting to you and you’d like me to give a version of it somewhere, let me know! The AI one will already have a rerun at KI Navigator in a few months, on home turf.
I still have some talks I’ll be giving this year, so stay tuned.
PS: if you wonder about the slide styling and the cyberwitcher stuff, I have a sizable announcement planned for that as well!