2 ms·
there is RL for LLMs which actually changes the weights but its more specialization than learning and wont counter the probabalistic nature of the thing
by saidnooneever 3mo ago
there is RL for LLMs which actually changes the weights but its more specialization than learning and wont counter the probabalistic nature of the thing
- SwellJoe 3mo agoTraining is not something you can just bolt on, and it generally requires even larger hardware than inference for a given model, and a huge amount of time. If you're aiming for "free", a OP claims, RL aint it.