3 ms·
It seems to me that in principle it should be possible to maintain attributions through the training process, so that Copilot outputs could come with a list of
by layer8 4y ago
It seems to me that in principle it should be possible to maintain attributions through the training process, so that Copilot outputs could come with a list of weighted sources, possibly discarding those that fall below a certain weight threshold. Doing so would likely be much more expensive in terms of the computational power needed for training, and probably also in the size of the model. But it would be great to actually be able to see what went into a specific Copilot output.