Securing Sanity: Gleen AI's $4.9M Mission Against AI Hallucinations

0:00the first thing i want to cover here is that as companies globally really accelerate the adoption of generative AI in various aspects of generative AI in various aspects of generative AI It's raised$4.9 million in a funding round that was led by Slow Ventures, Sixth Man Ventures, South Park Commons, and Spartan Group. And the round also featured participation from angel investors like Sam Lesson, who is the former VP of product management at Facebook or Meta. And Glean AI's primary offering is an anti-hallucination data layer software that's targeted initially at optimizing AI models for customer support roles.

0:58So Ashu Duby, who's the CEO and co-founder of Glean AI, revealed that and said essentially, quote, what we do is when we send data from a user to a large language model, we give facts that can create a good answer. If we don't believe we have good enough facts, we don't send the data to the LLM. So let's talk about some of the challenges with AI hallucinations. So generative AI tools, right, we've got a bunch of prominent LLMs, we've got ChaiGPT, Claude2, Llama2, Bard, all of these are essentially designed to generate responses based on human prompts. However, the reliability of the outputs, as we all know, can sometimes be quite questionable.

1:41So for instance, ChaiGPT once incorrectly answered a question about Earth eclipsing Mars, and it was delivering a confident but very fundamentally wrong response. So these kind of inaccuracies, they might be funny when you're using it yourself or casual interactions, but when we kind of look at how these affect enterprises, they pose a fairly significant risk, especially in sectors that deal with critical information. So we have healthcare, we have the financial industry. There's a lot of different industries where you just don't want to get this kind of stuff wrong and you really need things to be 100 % accurate, 100 % perfect, you know, you don't really have a margin of error on some of these industries, especially when it comes to the military, healthcare, financial services, all sorts of areas, right?

2:27So that's really where Glean AI's solution comes in and, you know, what their whole concept around accuracy is. So Glean has essentially developed a proprietary AI and machine learning layer that functions independently of any LLM. An enterprise can layer scans an organization's internal data, converting it into a vector database to enhance the functionality of the AI model's responses. So specifically, the layer performs several key functions. Number one, aggregates both structured and unstructured knowledge from various internal sources. Number two, it extracts key facts while reducing noise and redundancy.

3:09So Doobie actually noted that this process, quote, allows us to glean the signal from the noise, which is, you know, That's, I guess, what inspired the company's name there, Glean. The third thing it does is it then goes and constructs a knowledge graph to identify relationships between different data entities and to kind of, you know, retrieve the most pertinent information when someone is querying it. Then finally, what it does is it actually checks the LLM's output against the database of curated facts, opting for a cautious, you know, like, I don't know approach when evidence is lacking. and it does this to really kind of minimize the risk of hallucination.

3:48So in essence, Glean's AI layer serves as a quality control manager for chatbots, only enabling the LLM to respond when there is, you know, really high confidence in the comprehensiveness of the facts. And so the technology also allows users to quickly deploy custom support chatbots that can be customized to suit different use cases. So let's talk about the flexibility and, you know, essentially the early adoption of Glean, what we're seeing right now in the industry. So Glean AI's solution is model agnostic. So that means that it's supporting any LLM with API integrations. So while it supports popular models like, you know, OpenAI's GPT 3.5 Turbo, it also accommodates security conscious customers by offering some private server options, which I think, you know, when you're focusing on the enterprise, when this is the big customer you serve, you absolutely have to offer these private server options, essentially allowing companies to know that nothing, none of their personal data is ever leaving a private server, nothing is ever going to be used to train another AI model, and that they own everything that is, you know, everything being used, everything, all the data being given them, specifically, it's the data, right?

4:58Specifically, these enterprises do not want their personal data or information about their personal data to get leaked and cause a big headache or worse, right? Proprietary data get in the hands of a competitor or someone else, right? So that's the big worry. And so you really have to focus on these private servers. The early signs I would say indicate that the market is fairly receptive. Glean AI has already seen uptake in sectors that demand high levels of accuracy, such as quantum computing and cryptocurrency. So Estevan Villar, who, you know, communities, he works on community support at Matterlabs, praised the ease of Glenn AI's implementations, stating, quote, implementing Glean AI was close to no effort on our side.

5:40We just provided a few links and the rest was smooth. So as the enterprise adoption of LLMs continues to gain momentum, Glean AI aims to be on the forefront of ensuring these technologies are as accurate as they are versatile. They said, quote, our vision is every company will have an AI assistant powered by their own proprietary knowledge graph this vector database will become as important of an asset as their website enabling personalized automation across the entire customer life cycle so i personally have a obviously i'm very very bullish on ai and what's going on in the industry right now i think it's gonna be interesting to see um if indeed every company does have an ai chatbot that is you know the their whole um their whole knowledge base in my opinion i think this would be a really smart move this would be in the this would be going in the right direction and then And essentially, if you want information from a company, you go and query their knowledge base.

6:34And I think this will be really interesting to see how this scales in the future. But I think this is the direction we're going. And I think Galeen is setting itself up to be a really solid player, making sure that all of the information in these AI chatbots is 100 % accurate.