Notes

2025-05-04. Objectives Implicit in AI Reasoning Work.

AI companies and scientists define “AI reasoning capabilities” loosely, so it is unclear what it means for an approach to be better than another. I have noticed at least four distinct objectives implicit in AI reasoning work: (1) increase accuracy; (2) increase trust; (3) increase generalization; and (4) satiate curiosity. Each prima facie has different technical requirements and evaluation criteria.

2025-04-09. Notes on 100x (vs <2x) Research Opportunities.

I attended two Artificial Intelligence (AI) workshops to understand which applied statistics, causal inference, and machine learning contributions would accelerate the pace of innovation and scale of impact. I documented opportunities based on the workshops and my own experiences: (1) automatic, scalable custom measurement; (2) better AI agents; (3) technology to support compliance and privacy requirements; (4) patents and regulatory frameworks.

2024-12-07. Dynamic Local Average Treatment Effects.

These are (15 min) talk slides for my paper: Dynamic Local Average Treatment Effects.

2024-09-27. One Conda/Mamba Environment Can Support Both Python and R.

Scientists grok Python and R because the languages have complementary strengths, but many are unaware that both languages can be supported in a single virtual environment. Many are also shocked to learn that virtual environments support R. This is a write up of what I have verbally communicated over the years.

2024-11-22. Converting ipynb To and From Rmd.

Some people prototype exclusively in either Jupyter Lab or RStudio. During each prototype hand off, this means someone is responsible for converting ipynb files to Rmd (usually the receiver). Preferences do not have to slow progress :).

2024-06-04. AI Powered Feature Engineering for Machine Learning Models.

This is a slide deck summarizing a longer project proposal that I wrote. Others may find it useful (e.g. the evaluation slide).