thanhdotr

7 ms·

1. 1.▲
   by thanhdotr 1y ago

   is it unreasonable to expect some kind of good enough baseline guided by the prior of these llm to be the standard ? Should google's priors (or their training dataset and recipe) be allowed to guide societal's prior? same problem,
2. 2.▲
   by thanhdotr 2y ago

   well, as yann lecun said : "Adversarial training, RLHF, and input-space contrastive methods have limited performance. Why? Because input spaces are BIG . There are just too many ways to be wrong" [1] A way to solve the problem is
3. 3.▲
   by thanhdotr 4y ago

   Can you elaborate your point? "This is just a man appealing to pleas so as to ignore the roaring fire in front of him."