The Counterfactual

What is this about?

Topics will be fairly wide-ranging, but there’ll be an emphasis on Cognitive Science and Large Language Models (LLMs).

For representative posts, check out:

• So you want to be an LLM-ologist?

• Do large language models have a “theory of mind”?

• Mechanistic interpretability for LLMs, explained.

• What we talk about when we talk about LLMs.

Why “the counterfactual”?

One of the themes I’ll be exploring in this newsletter is the notion of thinking with baselines: given a claim—a number, a statistic—what should we compare it to? I think this connects more generally to counterfactual reasoning: what might have happened otherwise? I find both ways of thinking really helpful for enumerating a space of possible arguments or beliefs.

Does this cost money?

My goal is to make all posts freely available. However, a paid subscription gives you the option to vote in regular polls, which determine which topics I will write about for future posts. Past polls had led to various posts, such as an explainer on vision-language models and an empirical study investigating whether LLMs can modify the readability of a piece of text.