What do y’all think about a competition for the most capable 1000 parameter model?
The idea is that by having a competition with a prize, you incentivize the growth of that tech. In this case the tech would be making models as efficient as possible. A 1000 parameter limitation would get the job done.
OpenAI should put on a daily competition for the best tiny model (1k parameters). We should allow for mathematical functions or any type of expressions for each parameter, not just floating point numbers.
No rules on how the model is constructed, like through training, direct model weight generation or modification, etc.