2/2.

Energy Usage
The energy consumption required for training these models is a huge problem, but that’s a sunk cost. The decision to use AI involves weighing the environmental costs against the benefits. It is not my place to make this tradeoff for every user. Each individual or organization must evaluate whether the productivity gains and other advantages justify the energy expenditure.

Usage and Practicality
Users are going to use LLMs to help them use a terminal emulator whether support is built in or not. By building it in, the tool is more useful for these users. LLMs make mistakes all the time, but in my experience the quality is good enough to be useful.

1/2. AI features in iTerm2 caused some saltiness over on lobste.rs so I’ll put my comments here to reach a wider audience:

Opt-in
The AI features are entirely opt-in. If you have ethical considerations regarding the use of AI, you do not need to enable these features. Don’t provide an API key, and the AI functionalities will remain inactive. I don’t have VC funding and I’m definitely not footing the bill for you :)

Copyright Concerns
The issue of copyright in the context of AI, especially for code generation, is complicated. I think the principles of the GPL align more with using code for model training purposes, but I’m not a lawyer and I could be convinced otherwise.

Correct. All the AI features require an api key.

RE: https://curmudgeon.cafe/@StevenJV/112479395805377436