Opinionated articles on AI tools, research workflows, teaching, and productivity for academics and students.
In my process-and-trace talks I run three commercial AI detectors live on my own writing. They do well on text that is wholly human or wholly AI. On everything in between they disagree, and even when they "guess" right, they are answering the wrong question.
Our new article in The Conversation argues that AI is trained on the finished traces of human thinking rather than the process that produced them. That gap limits what machines can do, and it changes what happens to us when we hand them the thinking.
The ideas are still yours. So is the theorising, and most of the sentences. The writing drifts anyway. Marginalia learns how you write from your own published work, then flags where a draft stopped sounding like you.
Agentic AI won't out-predict a tuned gradient booster on large, clean data. Its value is in choosing and building the method, which sometimes means picking the booster.
Urika is an open-source multi-agent platform that automates exploratory data analysis. You bring a dataset and a question, and a team of AI agents does the rest.
Voice-to-text isn't new. Developers have used it for years, and academics know Whisper for transcription, yet most people still type their daily work. Most local tools keep your audio private and cost $50 or less, and this post lists the ones I use and recommend.
Cloud AI is convenient, until you can't use it. Student data, unpublished research, industry IP, and institutional compliance all push toward running LLMs on your own hardware, and open models are now capable enough to do it.
Universities panicked when ChatGPT arrived and reached for bans and detection tools out of fear of cheating. The bigger risk was failing to teach students how to use AI well. We went from cautious observers to building an entire course around AI-assisted learning.
Many of my colleagues and students freeze the moment I open a terminal. The command line looks hard because it's unfamiliar. AI assistants work in text, so avoiding the terminal now cuts you off from the AI tools that run there.
I stopped using PowerPoint, Keynote, and Google Slides this semester and now code my lecture slides in HTML with an AI assistant. It's faster and more enjoyable, and this post shows how to make the switch.
Most people use AI like a search engine, typing one vague question and getting one disappointing answer. Better results come from better input, and learning to prompt well improves every task you hand to these tools.
I barely use Word, Excel, PowerPoint, SPSS, or R Studio anymore. VS Code with AI assistants now handles my writing, analysis, slides, and tool-building in one place, and I work much faster for it.
Why I started this site, how AI tools transformed my academic workflow, and why I think every researcher and student can do the same, regardless of their technical background.
