The AI assistant complaints track with what I see on my end. Any general model I throw Wolfram Language at does noticeably worse than it does on Python. That part isn't surprising. There just isn't much public Wolfram code to learn from next to the mountain of Python sitting on github. It keeps guessing function names that sound plausible but don't exist. Spent an afternoon last week fixing hallucinated options on an NDSolve call it gave me.
What I'm curious about is whether this is actually on-device. Apple's framework caps local models around 3B params last I looked, and Claude is way bigger than that. So either there's some hybrid setup I haven't seen documented, or this is mostly a Claude SDK in FM clothing. Anyone tried it on a plane?
The anchoring thing is what gets me. Once I've seen the AI's first try, even when it's wrong, I can't really write fresh in my head anymore. I end up editing instead of starting over. Code quality usually ends up fine. Time-wise it's a wash or worse, you just don't feel it until you look at the clock at end of day.