Guardian agents: New approach could reduce AI hallucinations to below 1%
1 min read
Summary
Vectara has created a new service called the Vectara Hallucination Corrector, which reduces hallucinations in AI models.
The firm’s approach includes guardian agent software components that monitor and protect AI workflows.
Vectara aims to correct errors in workflows generated by LLMs while providing explanations for any changes made.
The system reduces hallucination rates for smaller language models under 7 billion parameters, to less than 1%.
Alongside the Vectara Hallucination Corrector, the company is releasing HCMBench, an open-source evaluation tool that helps to evaluate hallucination correction models.
The offering could enable enterprises to deploy AI in previously restricted use cases while maintaining accuracy standards.