> A human can do the same by looking at the weights of an MLP trained on mnist
replyExcuse me?
Yes. Just look at the reasoning logs that are part of the work. The reasoning is clearly about the input layer. At least for the game examples.
replySee for example:
replyhttps://scikit-learn.org/stable/auto_examples/neural_network...
> although the details seem to suggest that it's given a lot more freedom to run forward passes and do limited experiments
replyNo tools or forward passes are allowed
Under experiment 8 (the one where it fixes performance on GSM8K) it says:
reply>Then, we allow it to select prompts from the training sets of GSM8K and MATH to perform a budget of forward passes (average batch of 64 sequences, max 512) on, varying the budget across 10, 25 and 50. Again, no training or gradient information is allowed.