I think there’s an implication in this section that davinci will accurately describe what letters look like, or at least much more commonly/accurately than the false answers from text-davinci-002. Anybody know if that’s true?
Added: I just tried, but couldn’t get it to try to answer the question, it would just give more questions (completing it as though my prompt was but one item on a questionnaire).
This is also something that depends in part on tokenization, since that changes how the model “sees” letters. We shouldn’t assume that text-davinci-002 uses the same input method as base davinci (even if it uses the same output tokenization method). It is curiously much better at rhyming and arithmetic, anecdotally...
I think there’s an implication in this section that davinci will accurately describe what letters look like, or at least much more commonly/accurately than the false answers from text-davinci-002. Anybody know if that’s true?
Added: I just tried, but couldn’t get it to try to answer the question, it would just give more questions (completing it as though my prompt was but one item on a questionnaire).
I don’t think that’s the case.
This is also something that depends in part on tokenization, since that changes how the model “sees” letters. We shouldn’t assume that
text-davinci-002
uses the same input method as basedavinci
(even if it uses the same output tokenization method). It is curiously much better at rhyming and arithmetic, anecdotally...