what’s the security model around all these LLM tools that you have access to you code base, shell history, etc

Building @warpcast and @farcaster, new dad, like making food

Personally only give access to code that is open source. Services like ChatGPT do have team plans with the “promise” that context shares will be private and not used as data to train on, but still kinda sketch. 

I don’t give AI access to my shell. Only exception is Zed can read your terminal output but that’s on a per session and request basis.

It’s the “trust me bro” model if I’m not mistaken

to the large part benefits outweigh the potential risks today?

github lets you prevent sharing your snippets for training

E = T / A!

(The only security in software businesses in the future is dev velocity)