Their discussion covers:
- Amit shares One AI’s founding story and its mission
- We discuss how One AI is leveraging MongoDB’s developer data platform
- Insight into the cloud platform on which One AI runs
- Amit shares his reaction to the news of MongoDBs initiative with Google Cloud, to help developers build AI-powered apps
You can watch the full video here:
You can listen to the conversation here:
Disclaimer: The Six Five webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.
Patrick Moorhead: Hi, this is Pat Moorhead and the Six Five is live at MongoDB Local here in New York City, 2023. And as you can hear from the background, it is happening at this developer-focused event. Dan, how are you, my friend?
Daniel Newman: Yeah, it’s a great day, Pat. Anytime you can tie together big secular trends with the people that are making it happen, you’ve piqued my interest. We spend a lot of time focused on the board level, the C-level, but underneath all that, this app economy and now this AI economy that we’re in is being fueled by technologists, developers, cloud experts, and this place is absolutely full of them.
Patrick Moorhead: Yeah, it’s crazy how much time. It’s funny, I think I’ve spent 90% of my advisory time on AI.
Daniel Newman: Well, just that?
Patrick Moorhead: I think it’s a good… Well, last four or five months, but hey… Without further ado, let’s introduce our guest, Amit, from OpenAI. How are you?
Amit Ben: Hey, I’m good. Thanks for having me, guys.
Patrick Moorhead: Thanks for coming on the show. First time Six Five, appreciate that.
Amit Ben: Thanks for having me, guys, on the show.
Daniel Newman: Hey, very interested, Amit, in learning about your business One AI, so let’s start there. You’ve got to be on the right trend line at the right time, in the right market. Tell us about the business. Tell us. Give us the backstory. I always love hearing from founders about how it got started and what was the opportunity, the mission? What is the thing you think you’re solving for everybody?
Amit Ben: You asked for the backstory. I’ll tell you that, first of all, One AI, as a team, it’s our third AI company. We haven’t jumped on this bandwagon in the last four months. We’ve been in this space. Personally, I’ve been in it for 20 years. We started One AI about a year and a half ago. Our mission was at the time to bring language AI from the place it used to be where you had to… If you wanted language AI in your product, you had to bring all of these PhDs from Stanford and implement all of these research papers and hope to God something is going to come out of it. Our mission was, let’s take language AI. We saw it’s coming to the maturity point where it can already smell that it’s about to become mature enough to be able to be packaged and provided as a service for companies.
We started a company as an AI as a service company. It was a year and a half ago, and I think a lot of people told us we’re crazy and that AI is only something that you build in-house and only a few companies are going to be building it internally. But we took a bet and we think, we believe that language AI, like any other foundational technology, fundamental service, just like survey infrastructure and databases, is something that you have to specialize in. We specialize in language AI for businesses, for these business use cases where you really need the highest accuracy level and the best unit economics, where you need high accuracy, high performance, and good total cost of ownership, because that’s when you’re actually able to deploy AI into large production deployments, because the AI mandate is to save on costs, improve automation, improve scale.
If it takes too long to provide an outcome or costs too much per inference, you really didn’t do much. What we’re really seeing around the world in all of the emerging AI market is that the technology is coming to fruition and it’s emerging, to the sense where it’s front and center and top of mind for everybody, basically, but still to be able to deliver a lot of business centered use cases, which is where One AI is focusing on. Think about document analysis, customer service, sales interaction analysis, financial documents. All of these use cases that require very high accuracy levels and amazing unit economics because you have to deploy. You need to handle a million or 10 million interactions a month for every single use case. You really need to have your AI both scalable, performant, and accurate, at all at the same time.
Patrick Moorhead: Amit, I’m impressed on a lot of vectors. First of all, did I hear you correctly when you said the company was a year and a half old?
Amit Ben: Yeah.
Patrick Moorhead: It amazes me. I’m a seed investor and also some round A investing, and very rarely do I come in contact, let’s say, at a MongoDB big stage where I found a startup that is a year and a half old. Congratulations on that. It must be very unique and I need to do more research on the company. Given that we’re all here at MongoDB Local, I’m curious, what are you doing with MongoDB? Are you using their developer platform as maybe the basis for all the value add you put on top?
Amit Ben: As I said before, I believe in specialization and as an AI company, we want to make sure we only focus on building the best AI out there. But as a platform, we need to have a database, we need to store the data, we need to stream it, we need to process it. For that, we’re utilizing MongoDB and Atlas. When we started the company from the first day, we started building the foundation on top of Atlas. That allows us to completely check that box of the database and processing and storage foundations and to know that we’re future-proof and we’re not going to waste a single cycle on looking at anything else.
We all have been using MongoDB and Atlas in our previous roles in our previous companies, and we knew that’s going to be the platform. Today, all of our data is stored on MongoDB. We have huge amounts of different documents and different schemas and different parts of the systems. It’s actually schema-less with different types of documents. I think the huge power of MongoDB and where it really connects with language AI is because in One AI, we convert language, which is, by its definition, unstructured. We convert those unstructured data sources like conversations, emails, documents, into a structured document representation, which natively can very easily be stored on MongoDB. The fact that MongoDB has a dynamic schema allows us to continuously improve, innovate, and have dozens of different AI models generating all kinds of different data points, and store all of them into the database without even needing to mention that.
Daniel Newman: Sounds to me like there’s quite a bit of provenance within your team, the decision to run MongoDB. It sounds like this was something that, when you all came together, this wasn’t even a question.
Amit Ben: Yeah, yeah. I think we have a very long relationship with MongoDB. We had MongoDB in my previous company and a few of the development team that we brought on also used Atlas in their previous roles. It was a very natural choice for us to make.
Patrick Moorhead: Right.
Daniel Newman: Yeah. It seems that the company has really done a great job of engaging that developer community. While I think we’ll all agree that AI will shift what it means to be a developer, the developer will still be paramount. It’s just the way we develop. AI’s going to change it. You cannot possibly look at the what’s happening with the co-pilots and stuff and say it’s not going to change it, but just big data, just like low code, no code, it’s not going away. It’s going to get to be faster, more, all the things you talked about that we expect to be outcomes from AI.
All right, so enough trying to prognosticate what’s going to happen to developers. Of course, feel free to weigh in on that, but I want to weigh in another thing because NLP isn’t new. In fact, a lot of this AI stuff isn’t new. I keep saying I’ve been using generative in my Gmail for quite a long time. It’s been finishing my sentences for a few years now. Every so often, it’s actually better than I am. I’m like, “Wow, that that’s better. They say that better than I-”
Patrick Moorhead: I want it to just do half my email, though. Yeah.
Daniel Newman: But I’m saying search has been a very sophisticated AI for a long time. People are like, “This is new.” It’s not new. It’s like language NLP translation. I’ve used Google Translate around the world all the time. I use it because it’s the easiest tool, but you obviously see an opportunity with language that’s bigger than what is in market today. What was the demand? What is the thing that the market is asking for that you think you have a better mousetrap, you have a better story to solve as it comes to dealing with language in this AI era?
Amit Ben: Yeah, I think we’re all used to seeing AI in our day-to-day usage, but usually the AI that we’re using day-to-day is very simplistic in the types of prediction that it’s doing. It’s not actually creating added value. Translating something from one language to the other does not add or reduce information. It just translates it. I think the power of AI that is starting to be tapped today is to create value add capabilities. For example, if you had a sales conversation and you’re going to bring two sales guys, you’re going to ask them, was there a buy intent on a sales call? Good luck to getting them to agree on that, because that’s something that you really need to add in addition to what was said explicitly on the call. You want to add additional inference, additional insights. When you’re starting to talk about what is the value add capability, what were the objections on the sales call is a very loaded question.
There’s a lot of nuance and ambiguity that goes into analyzing that. The advent of the newest generation of AI allows us to make a lot more fine and granular analysis with a lot more nuance to the understanding. We’re not just converting what is or predicting the next word on an email, which is a relatively simple thing to do because, overall, we’re writing in a very repetitive fashion. We’re usually not writing completely novel stuff every day, just part of it might be changing. Most of it is templating. When you’re generating output, I know, for the companies that we’re working with, for example, when you’re trying to analyze financial documents, you’re trying to analyze sales conversations and interactions, you’re analyzing service interactions. You’re trying to engage deeper and to actually automate things that humans are normally automating or manually doing. You got to have a next level of understanding.
Patrick Moorhead: Yeah. Is there any low hanging fruit area? We talked a little bit about CSR, right? Customer service reps, we talked a little bit about sales conversations. That, by the way, could be part of a CSR conversation as well, but I’m also hearing about accounts payable, accounts receivable, certain areas. Where are you seeing some of the areas of heat and interest out there?
Amit Ben: I think I can bifurcate to two main streams. There’s the creative generative stream where app developers and companies are looking to leverage large language models to generate new stuff. That could be writing a story, an email, drafting something. That’s the generative stuff. These are usually the low hanging fruit because when you generate something, you’re at a much lower risk of making a mistake. You made something up anyway. If I phrase it differently or I put something in the email draft, someone’s going to correct it. I’m not too worried about accuracy and completion. When you write a story, you want the AI to be creative by definition. That’s where the low hanging fruits are, in the creative stuff. That’s easier to do.
When you’re looking at the other stream, from the extractive, either you generate or you extract. I believe that the most prolific type of data in the world is unstructured language data. It’s in our conversation right now. It’s in your sales calls. It’s in your internal company documents, in your emails, in your memos. It’s everywhere. Now, the ability to extract information from that unstructured data, which is completely taken over everything that we do, it’s in everything and everywhere. To be able to derive value from that, that’s where I think AI is really going to take off and deliver the next level of value outcome. Imagine our conversation right now converted to a structure that somebody can make sense of later on and make decisions based on, or even automate these decisions. That’s difficult to do.
Daniel Newman: Man, I really appreciate that because I think to your point, there’s a very spongy set of guardrails for most generative things. Did it generate? It’s like, was it good? Was it not? Other than factually, is it correct or incorrect, that’s where there’s a little bit more risk. But when it’s something like writing a sales email to somebody, it’s like, it can say something. You’re like, “It’s not really me, but it’s not wrong.” But when it comes to being able to… Because what you’re really doing is you’re blending generative with business intelligence.
Amit Ben: Yeah.
Daniel Newman: I have a three-day leadership meeting next week. I’d love to turn the mic on, record the whole thing, send it to One AI. You guys can tell me, after three days of conversation, what… I’d love to compare it to our typical process. Maybe we can call it the One AI challenge. We’re going to talk offline, but I do have to say thanks and let you go, because we do got to keep going on these conversations here today. It was really fun to talk to you.
Amit Ben: Thank you so much, guys, for having me on the show.
Patrick Moorhead: Thank you.
Daniel Newman: All right, everyone. You heard it there from One AI. We’re here at MongoDB Local, New York City, Patrick Moorhead and myself. We have a whole bunch more of these videos, so we hope you’ll hit that subscribe button. Stay with us all day here. Join us for all of our Six Five content on YouTube, on Twitter, LinkedIn, or wherever you have found this video. Pat, it’s time to say goodbye. See you guys on the next one.