../

Slides Preparation

Work done so far

  • AssertionBench explicitly states that it is not their intention to do any other metric other than proven %
  • But giving examples of HARM when those examples are pretty bad
  • They should have compared no-context learning against 1-shot and 5-shot