How can you possibly believe this? If this were true, why do open weight models think in real English tokens and not arbitrary characters?
How can you possibly believe this? If this were true, why do open weight models think in real English tokens and not arbitrary characters?