In a world where organizations are eager to integrate generative AI into their workflows, addressing the challenge of AI hallucinations has become paramount. Enter Gleen AI, a promising startup that claims to have the solution. Ashu Dubey, CEO and co-founder of Gleen, spoke exclusively to VentureBeat via video call, shedding light on their innovative approach.
Funding and Backing
Gleen AI has secured an impressive $4.9 million in funding from notable investors such as Slow Ventures, 6th Man Ventures, South Park Commons, Spartan Group, and angel investors, including former Facebook/Meta Platforms’ VP of product management Sam Lessin. This substantial backing will further the development of Gleen AI’s anti-hallucination data layer software, initially aimed at assisting enterprises in configuring AI models for customer support.
The Challenge of AI Hallucinations
Generative AI tools like ChatGPT, Claude 2, LLaMA 2, Bard, and others are designed to respond to human prompts and queries by generating data based on the input. However, these AI models often produce inaccurate or irrelevant information. For instance, ChatGPT’s response to “when has the Earth eclipsed Mars?” might sound convincing but is entirely inaccurate.
“What we do is, when we send data [from a user] to an LLM, we give facts that can create a good answer,” Dubey said. “If we don’t believe we have enough facts, we won’t send the data to the LLM.”
Gleen has developed a proprietary AI and machine learning layer that operates independently of the specific LLM chosen by their enterprise clients. This layer sifts through an enterprise’s internal data, transforming it into a vector database, and utilizes this data to enhance the accuracy of AI model responses.
Quality Control for AI
Gleen’s AI layer acts as a checkpoint, thoroughly validating the LLM’s responses before delivering them to end-users. This meticulous process eliminates the risk of chatbots providing false or fabricated information, ensuring data accuracy. Gleen’s software also empowers users to create customer-support chatbots with customizable personalities for various use cases.
Flexibility and Security
Gleen’s solution is AI model-agnostic, compatible with multiple leading models with API integrations. Whether clients prefer OpenAI’s GPT-3.5 Turbo model or a private server setup like LLaMA 2, Gleen accommodates various preferences.
For the most security-conscious customers, Gleen offers the option of using a proprietary LLM that remains isolated from the open internet. Ashu Dubey emphasizes that the root cause of AI hallucinations lies in insufficient relevant data, and Gleen’s accuracy layer addresses this issue by controlling input quality.
Seamless Integration
Implementing Gleen AI is a seamless process, as testified by Estevan Vilar, community support at Matter Labs. Gleen offers prospective customers a free “AI playground” to create custom chatbots using their company’s data. This innovative approach allows companies to harness the power of LLMs while ensuring the accuracy they and their customers demand.
The Future of AI Assistants
Gleen AI envisions a future where every company possesses an AI assistant fueled by its proprietary knowledge graph. This vector database will become as crucial as a company’s website, enabling personalized automation throughout the entire customer lifecycle.
