Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

This is a clever approach to reduce token usage. In my experience with Gemini 3 for code analysis, the biggest bottleneck isn't just the logic, but the verbosity of standard languages consuming the context window. A targeted intermediate language like this could make 'thinking' models much more efficient for complex tasks.


Almost like… a simplified set of instructions you would give a computer that get distilled down into machine code that executes on bare metal!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: