LLMs, Regexes, and Hospital Gowns

The latest two posts have been related to large language models.

The first post is a riff on my experience seeing whether an LLM could reproduce a regular expression I wrote for a client report this weekend.

The second post was on golden hospital gowns, a variation on The Emperor’s New Clothes in which the emperor is partially clothed in splendor and partially naked.

The two posts are related because the golden hospital gown metaphor applies to LLMs. My experience trying to prompt several LLMs to create my regex, and my experience trying to create a golden hospital gown image, were like most of my experiences with LLMs: an odd mixture of impressive and disappointing results.

I briefly mention cryptocurrency in my hospital gown post, risking the wrath of enthusiasts and detractors alike. My views on cryptocurrecy are complicated, not especially well informed, and in flux, and so I don’t write on the topic. In short, I’m somewhere between a fanboy and a hater, and I cannot convince myself to move entirely to either end of the spectrum. I find the space interesting to watch as a detached observer.