United states of america
OpenAI Proposes Confidence Thresholds to Combat ChatGPT Hallucinations
OpenAI's research highlights AI hallucinations in models like ChatGPT, proposing confidence thresholds to favor uncertainty over guesses, potentially reducing errors. However, this could make the tool less engaging by frequently admitting ignorance, risking user abandonment. The dilemma pits reliability against usability, urging industry-wide benchmark reforms.