Posts

Is GitHub down if the website loads but GitHub Actions doesn’t work? Technically, no. A lot of GitHub is still available. You can browse, review things, and push a commit. But if your release process depends on Actions, a big piece of GitHub is absolutely down for you. The same thing happens with clusters. The login nodes might be available while storage is having problems. Most of the compute nodes might be healthy while one GPU partition is offline. The owners can truthfully say that the cluster is up, while a researcher can say that they can’t do their work. ...

There is a lot of drama around OpenAI’s claimed solution to the Navier–Stokes existence and smoothness problem. I’m not going to try to evaluate the proof or untangle the dispute around it. One sentence near the bottom of OpenAI’s announcement was very interesting to me: While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models. Anyone using an AI tool for research stop and check their plan. ...

Stop adding things to Claude Code or Codex that you use once and then forget about. I like customizing my agents. Skills, plugins, extensions, whatever the particular tool calls them. But I want to combat the slow creep towards the dumb zone, and loading a bunch of unrelated stuff into every session works against that. Matt Pocock has a useful description of the smart zone and the dumb zone. Early in a session, the agent is at its best. The longer the conversation goes, the more mistakes start to pop up. This is where those early-days patterns of hallucinations really came from. The context window isn’t a promise that it’ll work well at all times. ...

In April 2025, I opened a very short issue in my PyLossless fork: there should be some sort of metric out there that we can add to the test suite to show the data is “cleaner” after pipeline runs Seemed reasonable and I figured that there would be a publication or something out there answering this. This came up because I was rounding out a test suite. I wanted something that could tell me if the data was “better” in a meaningful way, because I already had one making sure the input was different than the output. ...

In March 2025, I had an idea for improving quality control in PyLossless. I wanted to bring back being able to review an EEG recording to select independent components, choose a time window, and immediately compare the raw signal against the result of removing those components. It had been done in the previous version in MATLAB, but I was less familiar with the inner working of PyLossless at the time. I was also a complete novice when it came to things like Qt 5, PyQt, and the mne-qt-browser. ...

I recently gave a SHARCNET General Interest Webinar called “Writing More Code with AI Agents.” More than 300 people registered. Attendance was excellent, there were lots of questions, and I’ve had a decent number of follow-up conversations over email. Pretty happy with how it turned out! The response also confirmed why I wanted to give the talk. People are constantly asking me agents and what they should be doing with them. They’re watching other researchers and developers move very quickly with these tools, and there’s a real fear of missing out. They want to try them, but they don’t necessarily know where to start or how much of the output they should trust. ...

A MATLAB analysis pipeline can be battle-tested and still be difficult to share. The code can work, but the next platform may have the wrong MATLAB version, missing dependencies, or no MATLAB installation at all. I put together a small MATLAB container template to show that you don’t necessarily need to rewrite the analysis to solve this. MATLAB supports compiling a standalone application and packaging it as a Docker image. The deployed application runs with MATLAB Runtime rather than a full MATLAB installation. ...

A few years ago, I used Cookiecutter to show how to get from import blah to pip install blah. This sounds simple, but it isn’t. A Python script can live almost anywhere, but a formal package needs the right structure to be built into a wheel. It needs dependency metadata, versions, releases, a licence, and some way to test that it still works. Cookiecutter gave us a standard template. Instead of remembering every file and setting and creating them manually, we answered a few questions and started with a working package. I gave a talk about that workflow in 2022. ...

Starting an EEG study means making dozens of small decisions. Where will the original recordings live? How will they become a BIDS dataset? Which software environment will the study use? Where does preprocessing end and analysis begin? A student has to answer these questions before they’ve had enough experience to know what the consequences will be. Their supervisor then has to review those answers, remember what everyone decided, and figure out whether the project can still be reproduced two years later. ...

In April 2025, I wanted to put PyLossless on the DRAC clusters as an official module. I had at least half a dozen datasets to clean, and I was still managing many projects with separate virtual environments. Reinstalling the same preprocessing package for every dataset made no sense. It turned out that PyLossless was already available as a module. When I asked the internal team to update it, they gave me a list of changes they wanted first, especially if I planned to develop it myself. That request turned into active stewardship of the project and, eventually, co-ownership of its official repository. ...