it takes me longer to ask the LLM repeatedly to edit my file to the appropriate format than to just use regular expressions or other scripting methods myself
Not surprised. I would expect GPT to be better at helping me identify data cleaning issues, and helping me plan out how to safely fix each, and less good at actually producing cleaned data (which I wouldn’t trust to be hallucination-free anyway).
Maybe this is me being naive, but this seems like a topic where awareness of the destructive tendency can help defeat the destructive tendency. How about this, as a general policy: “I worry that this info will get misinterpreted, but here’s the full information along with a brief clarification of how I feel it should and shouldn’t be interpreted”?
To hostile listeners, you’ve given slightly less ammo than in the likely scenario where they caught you concealing the info. To less-hostile listeners, you’ve (a) built credibility by demonstrating that you’ll share info even when it doesn’t strengthen your cause, and (b) by explicitly calling out the potential misinterpretation you’re anticipating, you may make listeners more resilient against falling for that misinterpretation (inoculation / prebunking).
- By erring on the side of transparency while publicly acknowledging certain groups’ likelihood of coming to a distorted conclusion, I bet the CDC would have avoided a disastrous erosion of public trust and reinforcement of the “don’t trust the experts” vibe.
- By bringing up Bob’s evasive communication during the client prep and the anxiety it created for her, Alice would have deepened trust between them (granted, at the risk of straining the relationship if he did turn out to be irredeemably thin-skinned).
- …OK actually the cult/sect situation seems more complex, it seems to have more of the multipolar-trap (?) quality of “maybe no single individual feels safe/free to make the call that most people know would collectively be best for the group”.
It still seems to me that awareness of this trap/fallacy and its typical consequences can help a person or group make a much less fatal decision here.