A little parallel with Stories, isn’t it? The “brute force” look-up :) Otherwise, I don’t think it will that rapidly put a price tag on every problem to be solved, only on those that exist in the state space of reachable answers defined by the training data. We have a few more steps to massively invoke the idea of AGI being here. Effective learning on synthetic data, the same way we learn from our creations, the proper reward functions, and even the ability to reproduce and work collectively (already here at a certain level).
yes, a clear parallel! I would say synthetic data is everywhere now. o3 CoT training paths are supposedly fully synthetic. Nevertheless, I agree that we are still a few steps away.
A little parallel with Stories, isn’t it? The “brute force” look-up :) Otherwise, I don’t think it will that rapidly put a price tag on every problem to be solved, only on those that exist in the state space of reachable answers defined by the training data. We have a few more steps to massively invoke the idea of AGI being here. Effective learning on synthetic data, the same way we learn from our creations, the proper reward functions, and even the ability to reproduce and work collectively (already here at a certain level).
yes, a clear parallel! I would say synthetic data is everywhere now. o3 CoT training paths are supposedly fully synthetic. Nevertheless, I agree that we are still a few steps away.