mlpicker

6 ms·

1. 1.▲
   by mlpicker 3mo ago

   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 lear
2. 2.▲
   by mlpicker 3mo ago

   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 doc
3. 3.▲
   by mlpicker 4mo ago

   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. T