問題文
A team increased the size of their model to fix answers that were factually wrong about internal policies, but the problem persists. What does this indicate?
選択肢
- The temperature must be lowered further so the model stops varying its wording and settles on the most likely statement about the internal policy
- The content filter is interfering with the answers by removing the parts that quote the internal policy, leaving the model to fill the gap with its own wording
- The new model needs a higher tokens-per-minute quota, since a larger model consumes the budget faster and the truncated context is what causes the factual errors
- The problem is a lack of grounding in the internal documents, not a lack of model capability, so even a larger model would not know the policy either