I think the folks at Lander Analytics do really interesting cutting edge work and share a generous amount of what they learn. My own experience using a MacStudio for multiple side by side models is similar, but I need to look into these time series models more!

An AI harness for research. I have not tried it out because it seems far too optimized on writing papers which is not my primary output anymore.

I’m not surprised by the finding that even among AI users, code harnesses are not used. But if you are not finding value from AI for writing quantitative analysis code and research communication outputs its probably because you aren’t using a coding harness like OpenCode or Claude Code.

It’s important to remember that the most popular and accessed article about ChatGPT and student learning was retracted. The thing is, retraction is just a label, and people can choose to honor or not honor that label. So keep your eyes out for people citing the study.

Anil Dash is one of my favorite writers about technology. And this is why you will no longer see any Civilytics or Jared Knowles content on Twitter/X - I deleted all our posts after reading this article.

Simon Wilison has lots of great insights, and this one mirrors what I am finding. I am picking up a lot of “engineering” skills on the fly, while I am working on “vibe coding” things that I couldn’t deploy on my own in the past, but that I feel like I know very well what they should look and feel like.

Per Engzell and Nathan Wilmers provided some real insight on experiments with how well LLMs can produce quantitative social science paper production, even held back by using Stata of all things.

This is an interesting project I am tracking. There are so many AI agent harnesses, but harnesses make such a difference in the capabilities and usefulness of the tools. Particularly smaller LLMs really benefit from strong harnesses. Could be super useful!