It seems like we need a distributed training network where people can donate GPU time, but then any training result is automatically open.