Selected Talks

Like April 2011 Inspire9, Moonshine is a start… that works!

• You can use Moonshine to cast your linux desktop to a remote device
• It does automatic discovery and pairing
• It has tests! The Test Anything Protocol (TAP) is pretty awesome
• It responds to desktop resizing, and terminates sessions appropriately
• With the –tailscale option, it will work under strenuous network filtering

You’ll need to bring your own desk and chair or maybe sit on the floor… but it’s something with a past, a present and a future. Like April 2011 Inspire9.

When my friend Laura Summers invited me to co-author her PyCon DE talk about the Aesthetics of AI, we soon discovered that the collaboration had yielded content for 2 or 3 hours, maybe a full Kenneth Clarke-style TV series. We’d ended up with a common book from which different talks could be carved out by trimming some of the (let’s call them) “digressions”.

The linked video is Laura’s PyCon DE 2025 edit; it represents her personal curation of our common book. I think of it as Laura’s Kraftwerk-vs-Europop radio edit, though I was very happy that she added Franco to her slides about European dictators, and also kept in my meditations on Bullshit, Italian Futurism and the Munich 1937 Entartete Kunst exhibition.

One of these days we might collaborate again, this time towards my presentation of my own 50s-BeBop-vs-80s-Punk cut of our common material, which will no doubt also preserve a lot of Laura’s contributions.

Starting from first principles, Understanding Deep Learning builds an explanation of the mathematical workings of different deep neural network architectures, from the single-layer perceptron to cutting-edge topics like transformers and diffusion models.

Two chapters close the book with a change of pace and tone. “Ethics in deep learning” is presented as a literature review co-written with Philosophy Professor Travis LaCroix. “Why does deep learning work?” is titled as a question; Prince argues that it’s surprising that deep neural networks are easy to train and that they generalise.

Like the book, the talk touches on the “unreasonable effectiveness of deep learning”, including discussion of memorisation and generalisation, double descent and grokking, lottery tickets and pruning, and more.

Following previous PyCon AU talks like “Don’t Do This” (2013) and “Here be dragons: some elegant and ugly hacks in CPython” (2014), this talk grants some of the requests you never dared to make:

• PEP 336: Make None Callable — make your own None()! Easy in CPython with ctypes.
• The ?. operator for safe attribute access — add the syntax to PyPy, then use a source codec hack to rewrite it into legal Python 2.7.

The talk closes with a discussion of why extensible syntax as a standard Python feature might not be so crazy — it exists in other languages.

Repos: quiet_None · question_dot

Also on PyVideo.

This tutorial uses two sets of meta-programming techniques to derive Python code from semi-formal descriptions: first, to build Django models requiring very detailed validation supplied by the client in a spreadsheet; second, to build a Python API for a hardware device whose description is maintained by the engineers in a spreadsheet.

Business value: spreadsheet-driven meta-programming allows developers to respond to changes very fast by re-rendering Python code from modified spreadsheets, avoid transcription errors, save time, and clarify ambiguous specifications by making all programmer assumptions explicit in the code generator.

Technical topics covered: API design, Django modelling and validation, consuming spreadsheets via csv DictReaders, writing partially-applied functions via custom classes with a __call__ method, implementing fluent APIs with table-driven __getattr__ methods, implementing iterators with __iter__ and __next__, and more.

Also on PyVideo.

The thesis in eight minutes, through five examples:

1. When the Museum of Contemporary Art Barcelona opened, skateboarders had already claimed its plaza. Architect Richard Meier was not pleased. The museum, rather than fighting it, held a skateboarding exhibition in the space before the building even opened, and embraced the unintended use as part of its public identity.

2. Artist Andrew Shoben of Greyworld tuned the bars of a public fence in Paris so that running a stick along them plays “The Girl from Ipanema”. Installed without any authorisation, at the cost of carefully tuning every bar. In Japan, the same idea is deployed officially: road surfaces tuned so that driving at the correct speed makes the tyres play a pleasant melody. Virtuous behaviour as reward, encoded in infrastructure.

3. Self-styled “guerrilla architect” Santiago Cirugeda applied for council permission to place a skip (a building waste container) in a public street in Seville. The council approved. What showed up was a skip-shaped playground for children in a neighbourhood with nowhere to play. When officials objected, Cirugeda showed them the permit: nothing in the code said a skip had to be used for waste. The neighbours loved it, “it’s better than a skip!” they said. The funniest part of the intervention is that it was very cheap for the city: public money allocated to the arts was spent in the license for the skip, so the money returned to the city’s coffers.

4. The Enforcer is a roadside speed display that could be found on Punt road in Melbourne. It shows drivers their speed, but only when it’s under the limit; speeding drivers only see a “slow down” notice. I noticed something in myself when I saw it: every time we see a number, we read it as a score. The Enforcer turns compliant driving into a gamified experience rather than a chore.

5. A screen by the same stretch of road shows the suburb’s average household water consumption that week — 134 litres, or whatever it is. The nudge isn’t guilt about your own bill; it’s social comparison. Public policy people call these “nudges”: small design interventions that make the virtuous choice feel like the obvious one. The individual invoice tells you nothing useful: the single person who showers at the gym has a lower than average invoice, and the family with two kids uses more water than average. The neighbourhood display tells all of us the number to beat, which is a shared number.

Just as massively multiplayer games began hiring economists once their in-game economies became complex enough to study, we might eventually realise that public behaviour (how people use shared space, shared resources, shared infrastructure) follows the same logic as play. And if it does, we might want to hire game designers to work on it.

Coverage: Sydney Morning Herald