re: #52 goddamnedfrank

bsky.app

[Embedded content]

Yes, but if you were a first responder who wanted to use a large language model to answer this question, you could keep around a giant file of best practices for every king of emergency. You then feed this file into the LLM and ask your question stipulating that it only use the information you provided to answer it.
The better models will give you the correct answer 95%+ of the time, and if the answer doesn’t exist, it will say so.

This could be much faster than trying to find the right manual and page, or even querying a database.

It’s easy to break LLMs if that’s your goal. It’s easy to get garbage answers if you don’t know what you’re doing. It’s relatively easy to make really useful tools out of them if you know what you’re doing and are aware of the limitations. In this regard they’re no different than any other tool.