Generative Evaluator Tuning is a method that uses reinforcement learning to train e-LLMs by combining traditional training with rewards from KnowLLMs, which act as extra guidelines. If an e-LLM's output is logically incorrect according to KnowLLM or fails to meet specific criteria, it receives negative rewards, even if the output is similar to the ground truth based on similarity scores.

Authors

• Person: Not available Organization: arXiv
  Building Trustworthy NeuroSymbolic AI Systems - arXiv

Sources

• Building Trustworthy NeuroSymbolic AI Systems - arXiv arxiv.org arXiv via serper

Referenced by nodes (1)

• reinforcement learning concept