3 ms·
Good point. But many are already directly training on output from GPT. Probably more efficient than copying the raw training data. Especially if it relies on th
by bonzaidrinkingb 3y ago
Good point. But many are already directly training on output from GPT. Probably more efficient than copying the raw training data. Especially if it relies on this non-targeted approach.