PSA: this is *not* what any serious LLM-supported analysis would look like. Instead, LLMs are used to write code (that the researcher needs to understand!), producing reproducible analyses, or make judgments directly validated using human raters.
Gary Lupyan
A good cautionary tale for when someone claims that trusting data analysis done with an LLM is good because it’s “objective.”
In this example, Copilot apparently decided to ignore the data it was given and just run with all the subconscious biases its (human-made) training data inevitably gave it.
Katie Mack
"Based on the dataset you shared, US and UK responses differ mainly in tone, intensity, and wording style, even though they express similar emotional states"
New post on what happened when I got Copilot to do some data analysis: kucharski.substack.com/p/real-signa...
