Not sure who is right in the end, but all of these criticisms miss.

JGjoshuagrubbsphd.bsky.socialJan 28, 2024 3:35pm

You saw the headline, now look at what happened when scientists actually evaluate the headline

Check out my job market paper polisky! I use large language models to estimate ideal points. It allows you to separate policy and affect dimensions, and works with any corpus (voters, elites, rebel groups, short docs, long docs, etc.). No massive proprietary models (GPT) needed.

I work on methods in this area but am mostly disappointed with results. I find they are usually the wrong tool. Good for sure, but models hundreds of times smaller can do a lot of the same things without the baggage. Might change with some of the newer models coming out though!

@profmusgrave.bsky.social Please add me to polisky, thanks!