Head over to our on-demand library to view sessions from VB Transform 2023. Register Here
Before there was OpenAI’s ChatGPT, there was AI21 Labs’ Wordtune.
One of the first companies to bring generative AI tools into widespread commercial use, the Tel Aviv-based AI21 Labs came out of stealth in 2020, founded by AI and tech luminaries Amnon Shasuha (founder and CEO of Mobileye), Yoav Shoham (professor emeritus at Stanford University), and Ori Goshen (founder of CrowdX). AI21 Labs immediately launched Wordtune, an AI-based “writing companion that understands context and meaning” — which was a step beyond Grammarly and other spelling and grammar tools of the day.
While Wordtune immediately became popular — it made the list of Google’s favorite Chrome extensions in 2021 — when OpenAI’s ChatGPT debuted in November 2022 and quickly entered the cultural zeitgeist, it immediately sucked all of the air out of the generative AI room.
>>Follow VentureBeat’s ongoing generative AI coverage<<
VB Transform 2023 On-Demand
Did you miss a session from VB Transform 2023? Register to access the on-demand library for all of our featured sessions.
By 2023, AI21 Labs boasted some of the world’s most sophisticated large language models (LLMs) — its Jurassic-2 family of models consistently show up on “best of” lists, alongside those from OpenAI, Cohere, Google DeepMind and Anthropic. Jurassic-2 powers the company’s AI21 Studio, a developer platform for building custom text-based business applications, as well as Wordtune.
Still, AI21 Labs remains mostly known for Wordtune. And even with recent updates like Wordtune Spices, which cites its sources and can access the internet, and an upgrade last week that turned Wordtune into a generative AI platform, the product still lacks the chatbot interface that made ChatGPT so popular.
This has kept AI21 Labs flying under the radar, similar to Cohere in early 2023, as VentureBeat reported back in February. But the company’s mic drop of a funding round last week — $155 million, with participation from Nvidia and Google — makes it clear that as long as it is invited to the generative AI party, AI21 intends to show up and stand out, especially when it comes to competing with OpenAI for enterprise business investment.
VentureBeat spoke with co-founder Yoav Shoham after AI21 Labs’ funding announcement about the company’s goals, priorities and competition with OpenAI — especially given last week’s launch of OpenAI’s ChatGPT for Enterprise. This interview has been edited and condensed for clarity.
VentureBeat: A few months ago I wrote a story about Cohere flying under the radar when it comes to LLM companies. With the funding news yesterday, I was wondering if AI21 Labs has felt that way — that it has been a bit under the radar compared to, say, OpenAI or Anthropic.
Yoav Shoham: We’re so under the radar that you are even mispronouncing our name — it’s AI twenty-one.
VentureBeat: Oops, so it’s not AI two-one, it’s AI twenty-one. Sorry about that.
Shoham: AI for the 21st century. But you’re right. We’re way underbranded. And we like it fine that way. But now’s the time to change it. So it’s really a very opportune time. By year end, you’re going to see a big change in our presence and our branding. We’re in a situation now where insiders all know us. But you cannot write about AI without mentioning OpenAI. Most people mention us, but not everybody. That has to change.
VentureBeat: Why is this the time to change?
Shoham: We created AI21 Labs five years ago because while we saw, and we still see, amazing progress with deep learning applied to the language world, it is not sufficient. We are primarily an enterprise business, that’s the way we have looked at ourselves from the start. But it was also clear to us that the market wasn’t there yet. We created our own market with Wordtune. That’s just off the charts in terms of adoption, revenue and all that. Wordtune has been carrying the brand and doing great. But now, everybody’s experimenting with generative AI. This is the time to establish our brand presence, so that people know we have to be invited to the party. And when we’re invited to the party, we usually win — we jjust need to be invited. What we see in deals right now is OpenAI. We’re not really seeing anybody else. And we do well against them.
VentureBeat: Speaking of OpenAI, for months there have been so many companies creating generative AI for the enterprise. Cohere has their eye on the enterprise, Anthropic, and this week OpenAI came out with ChatGPT for Enterprise. And AI21 Labs does as well. What do you think of that target landscape?
Shoham: Enterprise businesses in general are where it’s at. This was the target for us from the beginning, and now it’s happening. Everybody is sprinkling it on their messaging and trying to target it. The question is, other than marketing, what do you have that makes your offering enterprise-ready? This is where one needs to understand the gap between exciting demos and robust systems. Language models are not like traditional software. They are expensive to run. They are inherently stochastic, unpredictable, unexplainable, and in the enterprise, if you’re brilliant 95% of the time, and garbage 5% of the time, you’re dead in the water.
So from the very beginning we put out effort into robustness, even at the expense of pizzazz. That’s been our whole philosophy in our own applications and it transfers to the enterprise, which needs robustness, reliability, predictability, and cleanability. That’s been our focus. We’re just out now with Jurassic v2, which is amazing, but what really sells are our task-specific models. We call these Lego pieces — all focusing on specific tasks, but still very general.
VentureBeat: Is there anything that you would say about ChatGPT for Enterprise, which was announced this week? How would you characterize that in comparison?
Shoham: It’s general purpose. ChatGPT is not a model. It is an application under which there is a model that calls for a chat modality. It is sometimes right, sometimes totally not what you want. Second, ChatGPT boils the ocean. They’re good boilers, they boil the ocean well, but it’s still boiling the ocean. As you want to hone in on a specific use case, you need to do a lot of work to get to the robustness, whereas we give you that capability almost off the shelf.
VentureBeat: When I talk to enterprises today, a lot of them are turning to open source models that they can fine tune on and feel like they have a lot of control over. How would you compare what AI21 offers to the open source options out there?
Shoham: The strongest open source offerings definitely trail behind the proprietary models, ours and OpenAI and Anthropic. The question is, does it matter? Sometimes not. But often, yes. I think the fact that you have access to the model weights gives you a sense of comfort. That data is warranted if you have engineers to actually go and look into the model. But usually it remains opaque and I can tell you that we at least can speak to our experience. I know there’s been a lot of buzz about open source models, that we have no moat, and all that. That’s not our experience. We have not found that we lose deals to open source. If we lose deals, it’s to OpenAI, mostly.
VentureBeat: And why do you lose those deals?
Shoham: It varies. I think sometimes it’s because of the brand. Nobody got fired for choosing IBM. And chat wasn’t front and center for our application, or for the applications of our customers. They were ahead of us there. It’s something that frankly, maybe we should prioritize that early on, and we’ll be coming out with that also, a Claude-like or ChatGPT-like thing. So to the extent that people want that, maybe a few times we lost deals that way.
VentureBeat: What would you say are your biggest priorities right now, as far as the business but also just the company generally? Are you planning on opening other offices? Are there other areas that you’re looking to expand?
Shoham: Our biggest challenge is managing increasd scale. We grew from 50 people to 100 people to 200. We’ll add another 100 people next year. You’re also going to see geographic expansion. That’s on the business side. Technology is an area that we have to run to stay in place. I personally am spending more and more of my time on the technology side. In two years, I don’t think we’ll be speaking about language models. We’ll be speaking about AI systems as part of language models. And our goal is to be at the forefront of that.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.