It is usually unhelpful to scream “wolf” on any occasion that state-of-the-art models make mistakes, but sometimes examples of extremely blatant failures are a reasonable corrective, alleviating ov...

Put on your philosopher’s funny-hat and prepare for a wild goose chase from inverse reinforcement learning, via evolutionary explanations, the language modeling objective, and Wittgenstein’s late p...

I remember vividly my first experimental project that involved preregistration. It felt crippling, intensely difficult. You have to plan in advance all the analyses that you will want to run on the...

This post describes how to integrate executable WebPPL code boxes into HTML documents, generated with Quarto. What’s WebPPL? WebPPL is a light-weight probabilistic programming language, which is ...

When testing the performance of Large Language Models (LLMs), a popular scoring function for multiple-choice options is referred to as “pointwise mutual information scoring” (“PMI scoring”) (e.g., ...

The goal of this post is to review a number of methods that approximate Bayes factors or marginalized likelihoods. Model comparison by Bayes factors A model \(M_i\), in the Bayesian sense, is a p...

Why this post? A number of papers recently crossed my path that reported on a computational model’s goodness-of-fit in terms of correlation scores, where the data to be explained were frequencies ...

Bundesagrarminister Christian Schmidt von der CSU geht der “veganen Wurst” an den Kragen. Fleischersatzprodukte mit Bezeichnungen wie etwa “Soya Schnitzel” oder “vegetarisches Döner” seien irreführ...

When Jones owns a dog, we usually say that he owns a dog, not a poodle (even if true and known) and not a pet. “Dog” is the basic-level category. If someone says that Jones owns a pet, we might i...

I taught a tutorial on Bayesian data analysis at KogWis-2016 today, which had a running example of a simple binomial coin flip scenario, by means of which I tried to highlight the conceptual differ...