3 ms·

being close is still impressive, especially for their first (released) model gives me hope that the training moat is even smaller than we thought