Posted 1/29/2024, 10:44:42 PM
Hugging Face Leaderboard Ranks AI Models on Tendency to Hallucinate
- New Hugging Face leaderboard ranks AI models on tendency to "hallucinate" or generate false information
- Tests models on factuality (contradicting facts) and faithfulness (deviating from instructions)
- Current top models are Meow, Stable Beluga, and Meta's LlaMA-2, with high scores indicating less hallucination
- Closed-source commercial models not yet tested for hallucinations
- Leaderboard aims to help identify most reliable, accurate models to mitigate AI misinformation