Earlier than there was OpenAI’s ChatGPT, there was AI21 Labs’ Wordtune.
One of many first firms to deliver generative AI instruments into widespread industrial use, the Tel Aviv-based AI21 Labs got here out of stealth in 2020, based by AI and tech luminaries Amnon Shasuha (founder and CEO of Mobileye), Yoav Shoham (professor emeritus at Stanford College), and Ori Goshen (founding father of CrowdX). AI21 Labs instantly launched Wordtune, an AI-based “writing companion that understands context and that means” — which was a step past Grammarly and different spelling and grammar instruments of the day.
Whereas Wordtune instantly grew to become widespread — it made the record of Google’s favourite Chrome extensions in 2021 — when OpenAI’s ChatGPT debuted in November 2022 and rapidly entered the cultural zeitgeist, it instantly sucked the entire air out of the generative AI room.
>>Comply with VentureBeat’s ongoing generative AI protection<<
By 2023, AI21 Labs boasted a number of the world’s most refined giant language fashions (LLMs) — its Jurassic-2 household of fashions constantly present up on “better of” lists, alongside these from OpenAI, Cohere, Google DeepMind and Anthropic. Jurassic-2 powers the corporate’s AI21 Studio, a developer platform for constructing customized text-based enterprise purposes, in addition to Wordtune.
Nonetheless, AI21 Labs stays largely identified for Wordtune. And even with current updates like Wordtune Spices, which cites its sources and might entry the web, and an improve final week that turned Wordtune right into a generative AI platform, the product nonetheless lacks the chatbot interface that made ChatGPT so widespread.
This has stored AI21 Labs flying below the radar, much like Cohere in early 2023, as VentureBeat reported again in February. However the firm’s mic drop of a funding spherical final week — $155 million, with participation from Nvidia and Google — makes it clear that so long as it’s invited to the generative AI celebration, AI21 intends to indicate up and stand out, particularly on the subject of competing with OpenAI for enterprise enterprise funding.
VentureBeat spoke with co-founder Yoav Shoham after AI21 Labs’ funding announcement in regards to the firm’s targets, priorities and competitors with OpenAI — particularly given final week’s launch of OpenAI’s ChatGPT for Enterprise. This interview has been edited and condensed for readability.
VentureBeat: Just a few months in the past I wrote a narrative about Cohere flying below the radar on the subject of LLM firms. With the funding information yesterday, I used to be questioning if AI21 Labs has felt that method — that it has been a bit below the radar in comparison with, say, OpenAI or Anthropic.
Yoav Shoham: We’re so below the radar that you’re even mispronouncing our identify — 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 twenty first century. However you’re proper. We’re method underbranded. And we prefer it high-quality that method. However now’s the time to alter it. So it’s actually a really opportune time. By 12 months finish, you’re going to see an enormous change in our presence and our branding. We’re in a scenario now the place insiders all know us. However you can’t write about AI with out mentioning OpenAI. Most individuals point out us, however not everyone. That has to alter.
VentureBeat: Why is that this the time to alter?
Shoham: We created AI21 Labs 5 years in the past as a result of whereas we noticed, and we nonetheless see, wonderful progress with deep studying utilized to the language world, it’s not ample. We’re primarily an enterprise enterprise, that’s the best way now we have checked out ourselves from the beginning. But it surely was additionally clear to us that the market wasn’t there but. We created our personal market with Wordtune. That’s simply off the charts by way of adoption, income and all that. Wordtune has been carrying the model and doing nice. However now, everyone’s experimenting with generative AI. That is the time to determine our model presence, so that folks know now we have to be invited to the celebration. And after we’re invited to the celebration, we normally win — we jjust should be invited. What we see in offers proper now’s OpenAI. We’re probably not seeing anyone else. And we do effectively in opposition to them.
VentureBeat: Talking of OpenAI, for months there have been so many firms creating generative AI for the enterprise. Cohere has their eye on the enterprise, Anthropic, and this week OpenAI got here out with ChatGPT for Enterprise. And AI21 Labs does as effectively. What do you consider that focus on panorama?
Shoham: Enterprise companies on the whole are the place it’s at. This was the goal for us from the start, and now it’s occurring. Everyone is sprinkling it on their messaging and attempting to focus on it. The query is, apart from advertising, what do you’ve got that makes your providing enterprise-ready? That is the place one wants to grasp the hole between thrilling demos and strong methods. Language fashions usually are not like conventional software program. They’re costly to run. They’re inherently stochastic, unpredictable, unexplainable, and within the enterprise, should you’re sensible 95% of the time, and rubbish 5% of the time, you’re lifeless within the water.
So from the very starting we put out effort into robustness, even on the expense of pizzazz. That’s been our complete philosophy in our personal purposes and it transfers to the enterprise, which wants robustness, reliability, predictability, and cleanability. That’s been our focus. We’re simply out now with Jurassic v2, which is wonderful, however what actually sells are our task-specific fashions. We name these Lego items — all specializing in particular duties, however nonetheless very normal.
VentureBeat: Is there something that you’d say about ChatGPT for Enterprise, which was introduced this week? How would you characterize that compared?
Shoham: It’s normal objective. ChatGPT just isn’t a mannequin. It’s an utility below which there’s a mannequin that requires a chat modality. It’s generally proper, generally completely not what you need. Second, ChatGPT boils the ocean. They’re good boilers, they boil the ocean effectively, but it surely’s nonetheless boiling the ocean. As you need to hone in on a particular use case, it’s worthwhile to do a number of work to get to the robustness, whereas we offer you that functionality virtually off the shelf.
VentureBeat: After I discuss to enterprises right now, a number of them are turning to open supply fashions that they’ll high-quality tune on and really feel like they’ve a number of management over. How would you examine what AI21 presents to the open supply choices on the market?
Shoham: The strongest open supply choices undoubtedly path behind the proprietary fashions, ours and OpenAI and Anthropic. The query is, does it matter? Generally not. However typically, sure. I believe the truth that you’ve got entry to the mannequin weights provides you a way of consolation. That information is warranted in case you have engineers to truly go and look into the mannequin. However normally it stays opaque and I can let you know that we at the very least can communicate to our expertise. I do know there’s been a number of buzz about open supply fashions, that now we have no moat, and all that. That’s not our expertise. We now have not discovered that we lose offers to open supply. If we lose offers, it’s to OpenAI, largely.
VentureBeat: And why do you lose these offers?
Shoham: It varies. I believe generally it’s due to the model. No person acquired fired for selecting IBM. And chat wasn’t entrance and middle for our utility, or for the purposes of our prospects. They have been forward of us there. It’s one thing that frankly, perhaps we must always prioritize that early on, and we’ll be popping out with that additionally, a Claude-like or ChatGPT-like factor. So to the extent that folks need that, perhaps a number of instances we misplaced offers that method.
VentureBeat: What would you say are your largest priorities proper now, so far as the enterprise but additionally simply the corporate usually? Are you planning on opening different places of work? Are there different areas that you just’re trying to broaden?
Shoham: Our largest problem is managing increasd scale. We grew from 50 individuals to 100 individuals to 200. We’ll add one other 100 individuals subsequent 12 months. You’re additionally going to see geographic enlargement. That’s on the enterprise facet. Know-how is an space that now we have to run to remain in place. I personally am spending increasingly more of my time on the expertise facet. In two years, I don’t assume we’ll be talking about language fashions. We’ll be talking about AI methods as a part of language fashions. And our purpose is to be on the forefront of that.