Whole Mars Catalog
@wholemars · Jun 29, 2026
As the model gets better at writing code, the usefulness of abstraction layers designed to make things easier for humans trends to 0
I would probably be concerned with memory safety, but the models are already very good at identifying and fixing buffer overflows etc
Elon Musk@elonmusk· Jun 29, 2026@aaronburnett Truly massive gains will come in ~3 months when the entire training and inference stack is written in C/C++ and massively simplified (most software layers will be deleted completely) and we exact-map Grok to work incredibly well on a GB300
Elon Musk
@elonmusk
With great power (direct memory management) comes great responsibility
10:20 AM · June 29, 2026 · 125.5K views
224
138
2.2K