3 ms·
It'd be interesting to see this retried with an open model so the standard and decensored model could be compared. That'd be a clue about whether the model is a
by roenxi 5mo ago
It'd be interesting to see this retried with an open model so the standard and decensored model could be compared. That'd be a clue about whether the model is avoiding it because it actively recognises the innuendo or if something else is going on.
- linhns 5mo agoWell then the picks will follow how the numbers are distributed in the training data. More popular numbers will show up more
- wongarsu 5mo agoThat's what you'd expect. But we don't know for sure why GPT4.1 chooses 69 only a quarter as often as a random dice roll would. And we don't know if this quirk is reverted by 'uncensoring' a trained model