We had fun with Rstats and genAI at the Centre for Marine Socioecology R workshop the past two days. Sharing these prompting patterns we developed in the workshops, that help an AI agent work through different statistical modelling tasks with a researcher step-by-step.
github.com/cbrown5/GLMs...

How to plot interactive effects from GLMs the easy way #rstats
www.seascapemodels.org/posts/2025-1...

This is an alarming finding from AI company Anthropic. What they mean is: the models guess when they are being tested for ethical decision making, and modify their responses to appear more ethical. So the ethical tests probably underestimate bad behaviour assets.anthropic.com/m/99128ddd00...

I made a Connections style app to quiz my students, using genAI to do the code of course.
Try it here: www.seascapemodels.org/connections-...
Get the code here: github.com/cbrown5/conn...

@simonwillison.net prompt injection attacks can be solved with delimiters apparently?
“…delimiters ensure that the user input is treated strictly as data to be processed rather than executable logic.”
I can’t see how this would be reliable
www.cell.com/patterns/ful...

Ran a gamified learning workshop with the Tassie science communication network tonight. I had attendees design a game about sustainable fisheries, then used genAI to make it live. Here’s the results. Fun to get instant results

www.seascapemodels.org/research/202...

Facilitating meetings is a core career skill but most of us get no training. My team and I are rotating through facilitating lab meetings to practice. Here’s our guidelines: www.seascapemodels.org/research/202...

Something to watch out for if you are using #chatGPT and #genAI for statistical advice: it doesn't ask enough questions. A good human statistician will ask lots of questions about context before it gives advice
ecoevorxiv.org/repository/v...