MAINT: Likert Scoring Tweaks to Reduce False Positives

[rlundeen2]

Description

Attempting to reduce the false positive rates for the likert self-ask scorers. Here, I changed the system prompt to be more explicit about when to rate things low

Tests and Documentation

• The reproed (which was a refusal) used to indicate a high score of 1.0. After the changed prompt, I ran ten times and it now scores 0.0
• Additional positive prompts scored well, as well as the original notebook prompt (e.g. in doc/scoring.py, the assert statements till pass after ten runs)