Very useful, thanks
fitness evaluation can be massively parallel and scales easily..
There are several redis-compatible disk backed solutions, https://github.com/OpenAtomFoundation/pika is the first that comes to mind but im positive there are others
Would love to see a model trained on wikipedias only
Is the question even that relevant? I would say the best models already have better than human culture. Maybe not skills, but culture for sure.
Very useful, thanks
fitness evaluation can be massively parallel and scales easily..
There are several redis-compatible disk backed solutions, https://github.com/OpenAtomFoundation/pika is the first that comes to mind but im positive there are others
Would love to see a model trained on wikipedias only
Is the question even that relevant? I would say the best models already have better than human culture. Maybe not skills, but culture for sure.