A new AI tool bills itself as a hallucination-free language model for enterprises.
Gyan, founded in 2018, works in a number of highly regulated industries, such as healthcare and the life sciences, where privacy and accuracy risks have significant implications. Gyan AI’s primary use cases include research and analysis or document review. It only processes text and is currently available in English.
The model operates in a fundamentally different way than mainstream LLMs. Put simply, it uses the rules of language, not math, to understand queries. And unlike others, Gyan AI is not pretrained on any data. Instead, clients dictate what materials—proprietary or external—the tool should draw on. “Gyan operates only from there and will make up nothing from anywhere else,” Joy Dasgupta, CEO of Gyan, told Fierce Healthcare. “When you don’t train your language model on data immediately, you have all these benefits.”
Those benefits, per Dasgupta, include: no hallucinations, no bias, no inconsistencies and no copyright infringement concerns, plus privacy and full traceability. Gyan AI is not exactly generative. It can produce reports or graphs without making up new words, while offering explanations of its underlying rationale.
“You can use your own corpus of information that you have that's, let's say, centric to your biotech, that you don’t want to share with anybody else,” Paramjit Kaur, a drug developer who serves as advisor at Gyan, told Fierce Healthcare. “I think for healthcare, that's really important, because you don’t want to share your secret sauce with everybody.” The model can be suited for a number of organizations, from providers to payers to regulatory agencies and beyond.
“We started Gyan with the firm belief that explainability of the results, transparent reasoning… tractable ways to improve model performance, and trust based on provenance, are essential for any foundational AI technology,” Gyan’s founder, Venkat Srinivasan, wrote on LinkedIn in early May. Srinivasan is a serial entrepreneur and founder of Innospark Ventures, an early-stage venture fund focused on AI startups.
Because Gyan AI runs in a client’s environment instead of on the cloud, it is priced differently than traditional LLM tools. For instance, there are no tokenization fees. This can ultimately make Gyan AI more affordable than other types of tools. And because Gyan AI is not transformer-based, like GPT, it does not require specialized, resource-intensive hardware such as Nvidia chips. As a result, Gyan AI runs on 200x less energy, per Dasgupta. For clients who care about sustainability, that is another meaningful draw.
Gyan’s core technical team consists of AI experts, computational linguists and engineers. The company has several patents and boasts that it excels in multiple benchmarks, including PubMedQA—a biomedical question answering dataset collected from PubMed abstracts—and MMLU, a popular benchmark for evaluating LLMs. Gyan AI ranks first, above GPT-4, Claude 3 and other models, on accuracy on both, according to data posted on Gyan’s website. “We’re on top and we’re not a black box,” Dasgupta said. Fierce Healthcare did not independently verify this data.
So, if this type of language model has so many benefits, why doesn’t everybody do it? “We’re puzzled,” Dasgupta said. “It seemed like the most obvious thing that you should have solved like 10 or 20 years ago, but it wasn’t,” Dasgupta said. The technical approach is, admittedly, a long road. That’s why Gyan spent five years perfecting the model. “It’s harder to build, but once built, it solves these problems,” Dasgupta said.
Gyan believes this type of model is the future of healthcare AI. Even with tools and use cases built for healthcare specifically, hallucinations, security and other concerns are often present in existing tools. As companies encounter these issues, there will be a growing demand for more reliable options.
“Healthcare today, budgets are being optimized, resources are stringent, time is always of the essence and of course, we don’t have any room for error,” Kaur said. The transparency of Gyan AI is “game-changing.”