Suppose this was true, an extremely small language model would outperform small language models. Sounds like inventing a compression which compresses any input by at least one byte.
Sorry for the snark, but the linked article doesn't explain why this obvious inference is false in this case.
1 comment
[ 2.7 ms ] story [ 15.1 ms ] threadSorry for the snark, but the linked article doesn't explain why this obvious inference is false in this case.