I find this project very interesting, and would like to try it out, but don't have the personal budget to do the training for my most-used model right now. Considering the relative stability of frontier coding models, would it not make sense to crowd-source optimized system prompts for common models? I know there are tradeoffs and non-determinism, in the process, but a library of standard prompts would lower the barrier to entry and could help make your excellent research more widely adopted.
I find this project very interesting, and would like to try it out, but don't have the personal budget to do the training for my most-used model right now. Considering the relative stability of frontier coding models, would it not make sense to crowd-source optimized system prompts for common models? I know there are tradeoffs and non-determinism, in the process, but a library of standard prompts would lower the barrier to entry and could help make your excellent research more widely adopted.