Post

Spent an hour yesterday trawling through my old replication packages with Claude Code making a skill so it could write R like me. Weird naming conventions, idiosyncratic formatting preferences, etc. Strongly recommend.

i think if you have a fun side project you want to test out, good use case to do it. E.g., "let's build a database that combines building permits and house price data"

I am skeptical of anything that lets the bot do more without more supervision. I feel like I do slower higher quality work with an LLM in chat mode, but the first pass is always wrong because of me or it. Letting it run wild for anything complicated seems like a recipe for disaster.

I tend to run analyses and make plots on my own without much aid. But data cleaning, harmonizing and munging is much faster with Claude, and that tends to be 90% of my job on a research project.

How long's the guest pass used for? I've mostly been using ChatGPT for workshopping ideas or understanding function calls - I usually don't let it integrate with my IDE; any advantages of Claude in this context?