3 ms·
Yeah not sure what’s impressive about this. Having the model be both the guesser and clue giver will of course have good results as it’s simply a reflections of
by ushiroda80 2y ago
Yeah not sure what’s impressive about this. Having the model be both the guesser and clue giver will of course have good results as it’s simply a reflections of o1’s weighting of tokens.
Interestingly this could be a way to potentially reverse engineer o1’s weightings