4 ms·

Here's the actual paper that was summarized in the OP: https://arxiv.org/pdf/2307.01850.pdf https://arxiv.org/pdf/2307.01850.pdf The paper is quite approachab

Here's the actual paper that was summarized in the OP: https://arxiv.org/pdf/2307.01850.pdf https://arxiv.org/pdf/2307.01850.pdf The paper is quite approachable and worth a quick scan. In essence, reusing output from previous generations results in rather wonky results which are clearly separable from the original training data.

lostmsu 3y ago
You probably have to clarify that it only applies to the direct transformer training with fixed weights for samples, and not RL variants or variable sample weights.