Dive Brief:
- Vectara unveiled a “guardian agent” that mitigates large language model accuracy Tuesday. The Hallucination Corrector tool identifies AI-generated inaccuracy, explains why it’s wrong and offers correction options.
- “While LLMs have recently made significant progress in addressing the issue of hallucinations, they still fall distressingly short of the standards for accuracy that are required in highly-regulated industries like financial services, healthcare, law and many others,” Vectara founder and CEO Amr Awadallah said in the release. Before Vectara, the executive founded Cloudera and had stints at Google and Yahoo.
- The company said its tool reduces hallucination rates to less than 1% for LLMs smaller than 7 billion parameters. Vectara also released a toolkit Tuesday to measure the performance of Hallucination Corrector.
Dive Insight:
Enterprises introducing AI into workflows know that the technology requires a new era of governance to maintain trust and ensure accuracy.
Companies like Ikea and American Honda have strengthened their governance playbooks in the past couple of years as AI roadmaps take shape. Incoming regulations, such as the European Union’s AI Act, are also requiring businesses to deploy guardrails as they move forward with adoption.
By next year, Gartner predicts enterprise spending will spike by more than 15% due to the amount of resources needed to secure AI, including access management and governance enforcement.
As AI agents gain enterprise traction, workers worry about the technology’s reliability, too. More than 1 in 3 employees said AI-produced work is not on par with their own, according to a Pegasystems analysis published in February.
Vendors have tried to bridge the gap between what enterprises need and what the tools can provide. Services that ground AI tools in enterprise data have proliferated, enabling better customization and more accurate results.
Smaller, task-specific models have also become more popular and are set to gain further adoption as CIOs look to minimize model errors. Gartner predicts enterprises will use small models three times more than general-purpose LLMs by 2027.
“The variety of tasks in business workflows and the need for greater accuracy are driving the shift towards specialized models fine-tuned on specific functions or domain data,” Sumit Agarwal, VP analyst at Gartner, said of the prediction in an April announcement.
Enterprise expectations of what vendors should provide will only continue. By 2028, 2 in 5 CIOs will demand vendors offer the capability to autonomously track, manage and contain the results of AI agent actions, according to Gartner.
