3 ms·
I didn't agree with the OP but just noting this is wrong: > These nets regurgitate the most average possible output given inputs. By definition. You seem to b
by nl 4y ago
I didn't agree with the OP but just noting this is wrong:
> These nets regurgitate the most average possible output given inputs. By definition.
You seem to be confusing the training objective with the capabilities.
For example:
> Why Can GPT Learn In-Context? Language Models Secretly Perform Gradient Descent as Meta-Optimizers
https://arxiv.org/abs/2212.10559 https://arxiv.org/abs/2212.10559