Seems like a reasonable trade-off to me. I'm happy for them to have a sustainable business model and people seem quite willing to pay monthly for AI. As long as they keep the free version and the ability to disable AI features then I think everyone wins.
Looks like modified placeholder addresses because the author didn't want to use real addresses. I don't think it could be used for internal routing since each octet is represented with a single byte (0-255) so having larger numbers for some internal routing would likely break the entire IP stack.
If the model is open source, I'm hoping for an option to be able to run this feature locally for free. They seem to have support for running other models locally (e.g. deepseek-r1 through Ollama), so I'm hoping they will keep that up with edit prediction.
The 6501 was pin compatible with the 6800, but had its own instruction set. After the lawsuit, the pins changed and we got the 6502. This meant that if you enabled 6800 support on the Apple 1 you could drop in a 6501 and use it with software made for the 6502. If you were really ambitious you could use a 6800 but then you'd have to write a new monitor ROM. There's no evidence that this was done at the time, but about 10 years ago someone did it: https://www.youtube.com/watch?v=ag6pWUhps7U