I chose a simple math equation that should almost assuredly be in a dataset of 1 trillion tokens exactly to check its basic pattern matching skills. The token "

I chose a simple math equation that should almost assuredly be in a dataset of 1 trillion tokens exactly to check its basic pattern matching skills. The token "10" and "20" are even in its hardcoded vocabulary set. I think GPT-4 might be exceeding HN discourse on ML at this point.