Making Large Language Models Work For You

A great talk from Simon Willison full of practical advice on large language models.

I liked his personal AI ethics:

• I won't publish anything that will take someone longer to read than it took me to write

• I never commit code if I couldn't both understand and explain every line

I also liked his point about how LLMs have helped him redefine “expertise”. Expertise isn't knowing every API of a tool by heart — that’s trivia. Expertise is knowing what the tool can do and what kinds of questions to ask in order to use the tool effectively.

Also this line just so accurately describes tech:

It's not easy to do well, but it's trivial to get up and running.

I also thought it was interesting that he suggests running the models your own laptop because you see a lot more of the hallucinations and holes in the models (vs. the ones from corporations which try to build in safeguards) which give you a better understanding of how they work.