Minimally Sufficient

LLMs make good classifiers but even better features

Real data can be so messy.. why not just use synthetic data!

I breakdown how to (invisibly) watermark a generative process

I developed a simulation-based win probability model for football.

How do we know whether a win probability is any good?

We dig into the details of the model:

It’s a lot; quite a heck of a lot

Or how to be the most effective Pokemon trainer

Noisy observations bias a whole distribution’s tails; SIMEX is an approach for correcting this

It gives arbitrary results when faced with dependent data

I've got a fever!

One thing that has always bugged me about Bradley Terry models is that it can't model rock-paper-scissors.

How to tell who's the best player?

Why get many samples when LLM do trick

I solve my guilty pleasure puzzle using Julia

How to correct prevalence estimates using PPI

Calibration is an important idea in statistical prediction.

I find a strange plot of p-values from a hypothesis test and investigate.

How quickly can I generate a Poisson(1) random variable?

I investigate SMBC’s novel correlation-chain methodology.

Sometimes you have to take risks if you want to win.

How do you sample from a stream of unknown length?

I apply a Bradley-Terry model to the NCAA basketball tournament.

An easy way to perform inference after model selection.