This technique is a much bigger deal than it seems
For a few $ you can fully replicate any product you have access to just by bruteforcing agents against it
We did this internally at Ordalie with microsoft word online to get pixel perfect docx rendering and it worked insanely well
It's about expressing the constraints and levers in the harness. agents excel at tasks they can autoverify -- expect much more of this!
Quote
Elliot Arledge
@elliotarledge
Introducing Netherite!
Minecraft 1.11.2, rewritten from scratch in C and CUDA, bit-verified against the real game.
This is one trained agent playing in it - then the renderer it actually trains through - then 7,200 live worlds stepping in lockstep on one GPU.
The media could not be played.
