"Gleen AI secures $4.9 million in funding to tackle AI illusion problem with enterprise data"

2023-09-06

Global organizations are seeking to rapidly assess, test, and deploy generative artificial intelligence (AI) into their workflows, whether in the backend, frontend (customer-facing), or a combination of both. Many decision-makers have ongoing concerns about several unresolved issues, including the problem of AI illusions.

However, a new startup called Gleen AI has emerged, claiming to "solve the illusion" problem.

Today, Gleen AI announced that it has raised $4.9 million in funding, with investors including Slow Ventures, 6th Man Ventures, South Park Commons, Spartan Group, and other venture capital firms and angel investors, including former Facebook/Meta platform product management vice president Sam Lessin. The funding will be used to continue developing enterprise-focused anti-illusion data layer software, with the initial goal of helping businesses configure AI models to provide customer support.

The Illusion Problem

The training purpose of generative AI tools, such as popular commercial large language models (LLMs) like ChatGPT, Claude 2, LLaMA 2, Bard, is to respond to user input prompts and queries by generating data related to the words and ideas inputted by the user.

However, generative AI models are not always accurate. In many cases, the generated information is incorrect or irrelevant, but it is a result of the model associating certain phrases from the training process with the user.

For example, ChatGPT attempted to answer the question "When did Earth surpass Mars?" and provided a convincing but completely inaccurate explanation (the premise of the question itself is incorrect and inaccurate - Earth cannot surpass Mars).

While these inaccurate responses may sometimes be humorous or entertaining, they can pose significant risks for businesses that rely on them to generate accurate data for employees or users, especially in highly regulated industries such as healthcare, pharmaceuticals, heavy industry, and life-or-death information.

Gleen's Solution

Dubey said, "When we send data from users to LLM, we provide facts that can generate accurate answers. If we believe the facts are insufficient, we won't send data to LLM."

Specifically, Gleen has created a proprietary AI and machine learning (ML) layer separate from any LLM that enterprise clients want to deploy.

This layer securely filters the enterprise's own internal data, transforms it into a vector database, and utilizes this data to improve the quality of AI model responses.

Gleen's layer performs the following functions:

Aggregates structured and unstructured enterprise knowledge from multiple sources such as help documents, FAQs, product specifications, manuals, wikis, forums, and past chat records.

Filters and extracts key facts, eliminating noise and redundancy. Dubey stated, "This way, we can get signals from noise." (This is also the origin of the name Gleen.)

Builds a knowledge graph to understand relationships between entities. This graph helps retrieve the most relevant facts for a given query.

Checks LLM responses based on the filtered facts before outputting. If there is insufficient evidence, the chatbot will say "I don't know" instead of risking providing incorrect or false information.

This AI layer acts as a quality control checkpoint for chatbots, allowing cross-checking of LLM responses before they are outputted to end users, eliminating the risk of chatbots providing erroneous or fabricated information.

Dubey explained, "We only use LLM when we have high confidence in the facts. Otherwise, we explicitly ask the user for more information."

Gleen's software also helps users quickly create customer support chatbots tailored to their customers and adjusts their "personality" based on use cases.

Gleen's solution is compatible with leading models that support application programming interfaces (APIs).

For customers who want to use the most popular LLMs, it supports OpenAI's GPT-3.5 Turbo model. For customers concerned about their data being sent to LLM hosting companies, it also supports LLaMA 2 running on private servers (although OpenAI has repeatedly stated that it does not collect or use customer data to train its models unless explicitly allowed by the customer).

For customers sensitive to security, Gleen also offers the option to use a proprietary LLM that never touches the open internet. However, Dubey believes that LLM itself is not the root cause of illusions.

Dubey said, "LLM creates illusions when there are not enough relevant facts to base responses on. Our accuracy layer solves this problem by controlling the input to LLM."

Preliminary Feedback is Encouraging

Currently, customers using Gleen can ultimately obtain a customized chatbot that can be integrated into their own Slack or serve as a support agent for end users.

Gleen AI has already been used in quantum computing, cryptocurrency, and other technology fields where accuracy is crucial.

Estevan Vilar, community supporter at Matter Labs, a company dedicated to making the cryptocurrency Ethereum more suitable for enterprises, stated, "Implementing Gleen AI was almost effortless on our end. We just needed to provide some links, and the rest went smoothly."

Gleen offers potential customers a free "AI playground" where they can create their own customized chatbots using their own company's data.

As more companies attempt to leverage the powerful capabilities of LLMs while mitigating their drawbacks, Gleen AI's accuracy layer can provide them with the accuracy needed to deploy generative AI.

Dubey stated, "Our vision is for every company to have an AI assistant supported by its own proprietary knowledge graph. This vector database will become as important an asset as their website, enabling personalized automation throughout the customer lifecycle."

RECENT AI NEWS

RECENT AI TOOLS