Vitalik Buterin
@vitalik.eth
Interesting OSI debate on how to define "open source AI": https://lwn.net/SubscriberLink/995159/a37fb9817a00ebcb/
11 replies
299 recasts
1004 reactions
keccak256🎩
@keccak256
I believe the data used to train needs to be available and the weights can be replicable in order to be open source. The analogue is that data is the source code. The transformer model is the hardware used to run the code. That's how I see it.
1 reply
0 recast
0 reaction
keccak256🎩
@keccak256
The model weights are the binaries.
0 reply
0 recast
0 reaction