But it’s interesting to consider whether the training algorithm allows, in theory, the model to learn enough of a token’s properties to answer any question about it /as if/ the model could read each character.
But it’s interesting to consider whether the training algorithm allows, in theory, the model to learn enough of a token’s properties to answer any question about it /as if/ the model could read each character.
When you ask “what is the capital of France” and ChatGPT says Paris [72872], it learned that. It didn’t look up the answer on a map.
So it could learn there are 2 b’s in blueberry. Even if it doesn’t count them.
When you ask “what is the capital of France” and ChatGPT says Paris [72872], it learned that. It didn’t look up the answer on a map.
So it could learn there are 2 b’s in blueberry. Even if it doesn’t count them.
I’m target market for the X Half but even for me it’s a little… too quirky 🙃
I’m target market for the X Half but even for me it’s a little… too quirky 🙃
So… unusual but not unprecedented.
So… unusual but not unprecedented.