Seth Meyer
@sjmeyer.bsky.social
Economist-in-training. BS Mathematics ('15), MA ('26) + PhD ('30) in economics pending! Believer in democracy through data. Unapologetic math geek.
Reposted by Seth Meyer
If you absolutely, positively must use an LLM for analysing data: get it to write code to do the analysis and then run the code.
Code can at least be read and verified, and the results reproduced by others. The raw output of an LLM cannot be trusted.
Code can at least be read and verified, and the results reproduced by others. The raw output of an LLM cannot be trusted.
Have deleted my earlier chart - clearly bad mistakes but would also like to set out what happened, as a cautionary tale. Although it makes me look extremely silly, it's worth sharing.
This specific doc? I count three mentions of “humber” in the industrial strategy doc, 24 mentions of “Yorkshire” with or without qualifiers, zero mentions of “yorkshire and humber”.
June 23, 2025 at 1:50 PM
If you absolutely, positively must use an LLM for analysing data: get it to write code to do the analysis and then run the code.
Code can at least be read and verified, and the results reproduced by others. The raw output of an LLM cannot be trusted.
Code can at least be read and verified, and the results reproduced by others. The raw output of an LLM cannot be trusted.
Researching differences in income due to evictions in racially segregated neighborhoods, à Milwaukee, WI? Better do a presentation on that! #EconSky
June 13, 2025 at 6:45 PM
Researching differences in income due to evictions in racially segregated neighborhoods, à Milwaukee, WI? Better do a presentation on that! #EconSky