In today’s fast-paced world, organizations are eager to incorporate generative AI into their workflows to optimize their backend and customer-facing operations. However, a common concern among decision-makers is the problem of AI hallucinations. These false or inaccurate responses generated by AI models can have significant consequences for businesses, particularly in highly-regulated industries like healthcare and heavy industry. Addressing this issue head-on, Gleen AI, a newly emerged startup, claims to have the solution to combat AI hallucinations. In an exclusive video call interview with VentureBeat, Ashu Dubey, the CEO and co-founder of Gleen, shed light on the disruptive technology the company has developed. Today, Gleen AI proudly announces a successful $4.9 million funding round from key investors, including Slow Ventures, 6th Man Ventures, South Park Commons, Spartan Group, and prominent angel investors such as Sam Lessin, former VP of Product Management at Facebook/Meta Platforms.
Generative AI tools, including popular large language models (LLMs) like ChatGPT, Claude 2, LLaMA 2, and Bard, are designed to generate responses based on user prompts and queries. However, these AI models are not always accurate and often produce irrelevant or erroneous information. Despite their advanced training, they can be prone to misconceptions and inaccuracies due to associations made during their training process. For instance, when asked about the Earth eclipsing Mars, ChatGPT provided an elaborate yet entirely inaccurate explanation. Such inaccuracies may be amusing in certain contexts, but they pose a grave risk for organizations relying on AI models for critical data in industries such as healthcare, medicine, and heavy industry.
Gleen AI has developed a revolutionary anti-hallucination data layer software to address the challenges posed by AI hallucinations. The startup’s primary focus is assisting enterprises in configuring AI models to provide customer support effectively. Gleen’s proprietary AI and machine learning (ML) layer operates independently of any specific LLM, allowing their enterprise customers to seamlessly integrate their preferred AI model. By sifting through an enterprise’s internal data, Gleen’s software transforms it into a vector database, enhancing the accuracy and relevance of AI model responses.
The core capabilities of Gleen’s data layer include aggregating structured and unstructured enterprise knowledge from multiple sources such as help documentation, FAQs, product specs, manuals, wikis, forums, and past chat logs. Gleen’s software curates and extracts the key facts while eliminating noise and redundancy, allowing it to “glean the signal from the noise.” These facts are then structured into a knowledge graph that facilitates understanding the relationships between different entities. This knowledge graph enables the software to retrieve the most relevant information for a given query. Before delivering the AI model’s response, Gleen’s layer cross-checks it against the curated facts to ensure accuracy. If sufficient evidence is lacking, the chatbot will candidly respond with “I don’t know,” preventing the risk of hallucination.
Gleen’s AI layer acts as a dependable checkpoint, verifying the accuracy of the AI model’s responses before they reach the end user. This effectively eliminates the possibility of the chatbot providing false or fabricated information. Dubey aptly compares Gleen’s role to that of a quality control manager for chatbots. The software only engages the AI model when it has extensive confidence in the comprehensiveness of the facts at hand. If additional information is required, the chatbot transparently communicates the need to the user.
Gleen AI’s solution is AI model-agnostic, meaning it can seamlessly integrate with any leading AI model that offers APIs. For customers seeking the most popular LLM, Gleen supports OpenAI’s GPT-3.5 Turbo model. Additionally, for security-sensitive customers, Gleen offers a proprietary LLM that never touches the open internet, ensuring data privacy. Dubey stresses that LLMs themselves are not inherently responsible for hallucination; rather, it is the lack of relevant facts that leads to flawed responses. Gleen’s accuracy layer effectively solves this issue by carefully controlling the inputs to the LLM.
Implementing Gleen AI is a seamless process for its customers. Companies such as Matter Labs, focused on making the cryptocurrency Ethereum more enterprise-friendly, have experienced minimal effort while integrating and utilizing Gleen’s software. Community Support at Matter Labs, Estevan Vilar, attests to the simplicity of implementation, stating that they only needed to provide a few links, and the rest was “smooth sailing.” Gleen AI offers potential customers a free “AI playground,” allowing them to harness their own company’s data to create a custom chatbot tailored to their specific use cases.
Looking towards the future, Gleen AI envisions every company having an AI assistant powered by its own proprietary knowledge graph. The vector database created by Gleen’s software will become a valuable asset, as crucial as a company’s website. This knowledge graph will enable personalized automation across the entire customer lifecycle, enhancing the efficiency and accuracy of AI-powered operations.
Gleen AI provides a groundbreaking solution to the problem of AI hallucinations. By leveraging an anti-hallucination data layer software, Gleen ensures that AI models deliver accurate and reliable responses to user queries. With a focus on improving customer support, Gleen’s software curates, extracts, and structures enterprise data to create a comprehensive knowledge graph. By cross-checking responses against curated facts, Gleen prevents false or fabricated information from reaching end users. As more companies seek to harness the power of AI models while mitigating the risks associated with AI hallucinations, Gleen AI’s accuracy layer offers a promising path forward. With its user-friendly implementation and support for various AI models, Gleen AI empowers companies with personalized automation, revolutionizing the way businesses interact with their customers.
Leave a Reply