RAISE Summit 2025 | Jonathan Siddharth, Turing

Clips
More from RAISE Summit 2025

Jonathan Siddharth

CEO

Turing

play_circle_outline Jonathan Siddharth Explores Turing's Pivotal Role in AI Data Generation for High-Quality Training of Frontier Models

play_circle_outline Turing benefits from the need for neutral data partnerships.

play_circle_outline Future of Work: How Automation and Advanced AI Capabilities Will Transform Computer-Based Tasks in Just Three Years

Info
Transcript

Jonathan Siddharth, Turing

Jonathan Siddharth

CEO Turing

In this RAISE Summit interview, Turing chief executive, Jonathan Siddharth, joins theCUBE’s John Furrier to reveal how his fast-growing AI infrastructure company is helping frontier labs and Fortune 500 enterprises move from proof-of-concept to production. Siddharth details Turing’s hybrid data approach that blends expert human insight with synthetic generation, powering model builders like OpenAI, Anthropic and Meta, while fine-tuning smaller domain-specific models for customers such as Disney and Johnson & Johnson.

The discussion explores Turing’s re... Read more

explore Keep Exploring

What does the company Agentic do in the field of AI infrastructure? add

What are the reasons clients are seeking to work with a neutral data partner like Turing instead of other labs? add

What advancements in automation are expected to occur in the workplace within the next three years? add

bolt Powered by CUBE AI

Jonathan Siddharth, Turing

search

>> Welcome back, everyone, to theCube's live coverage here in Paris, France for RAISE Summit 2025. I'm John Furrier, your host of theCUBE. Two days of wall-to-wall coverage, unpacking the innovation in AI, how enterprise of hyperscalers our developers and builders are leveraging the new technology, bringing new kinds of value, extract value. That's the focus of the event. Of course, we're bringing you all the action live. Jonathan Siddharth, CEO of Turing, and founder, great to have you on.

Jonathan Siddharth

>> Thanks for having me, John. - Yeah, great to have you on.

>> Again, one of the hottest areas is getting people up to speed. Skills gaps, change, management has always been kicked around, but more than ever, you've got technical platform changing, business logic evolving, and intersecting. It's causing people a lot of excitement, but they're looking for the confidence because they want to get their data to work. This is what you guys do. Explain what you guys do, because you guys are attacking this problem, or just share the problem statement and what you guys do.

Jonathan Siddharth

>> Yeah. Thank you, John, for having me. It's great to be here. So agentic is one of the fastest growing AI infrastructure companies in the world that's also profitable, and we do two things. On one hand, we generate data that powers all the leading frontier models, all of them. And we help the models get better at coding, reasoning, thinking, STEM, multimodality, multilinguality. We help the models become agentic, and we generate this really high quality training data to help the models improve. And on the second hand, we also fine tune models for the enterprise where we work with big Fortune 500 clients that have important business problems they want to solve. We help them build end-to- end AI systems on top of these models. And today, Turing works with companies like OpenAI, Anthropic, Meta, Google, NVIDIA, Microsoft, almost every lab you've heard of, and enterprises like Disney, Pepsi, Johnson & Johnson, Coinbase, all these large companies.

>> So you serve all data gatherers' needs. You serve those customers. You don't really have a . You're horizontal.

Jonathan Siddharth

>> We are horizontal. We serve all the model builders that are training these frontier models. We generate data for them. And we help enterprises that want to solve business problems with AI, where we build an end-to- end AI solution on top of the models.

>> Yeah. One of the things that's emerging, I want to get your thoughts on, if you don't mind, is vertical models. We hear a lot about vertical fine- tuned healthcare model, pre-baked, that's an example. And then the big models of solving all that called general purpose, multiple prompts, not just first prompt out. You got multi-step reasoning kicking in. What specifically are you guys doing with the models? Is it changing the data equation? Is it helping them do evaluation? Are you adding math and probability into it? What are some of the science behind what you do?

Jonathan Siddharth

>> Yeah, great question, John. So to unpack that a little bit, we have to go to what's changing in the world's data landscape today, right? As these models have gotten smarter and smarter, the data that is needed to improve the models has become increasingly harder to generate. The models are improving in depth, in breadth, and they're becoming agentic. Now, to solve these problems, you need a platform that has the sharpest human minds on it.

>> Yeah. - The human experts in coding, in math, in STEM,

Jonathan Siddharth

>> you want them on the platform working alongside AI to generate data. The platform itself has to be able to flex up and down to cope up with these unpredictable demands that the labs might have. And you also want to be able to create these RL gyms, these reinforcement learning gyms, to train agents, right? So the world has shifted where it used to be that the labs needed a data factory or a recruiting partner. Today, the labs need a research partner. And Turing is a research accelerator, where we partner with researchers at these frontier labs to, step one, evaluate the models, step two, identify gaps in the models where the model's performance could be improved, and step three, generate custom data to plug those gaps. And then we evaluate again. So it's evaluate, generate data, fine-tune the model, see if the performance improves.

>> So you're reinforcing the reasoning aspect of

>> what they think is the right answer. So you potentially are going to help solve the hallucination problem. Would that be accurate?

Jonathan Siddharth

>> We are helping the models become more and more accurate over time. And I think as they get more and more accurate, hallucinations will decrease. It's hard, in this particular architecture, for hallucination to completely go away, unfortunately.

>> Yeah. Yeah. - But we are making the

Jonathan Siddharth

>> models smarter and smarter.

>> And we cut our teeth by -- when we first partner with OpenAI on GPT-3 to teach GPT-3 to code and to do function calling. And then from there, we expanded to a variety of different offerings.

>> So are you generating real data, synthetic data?

>> What's the data generation strategy? Are you getting data for them? Are you data brokering, or are you just generating synthetic data?

Jonathan Siddharth

>> So we generate what I would call hybrid data, where it's a combination of human data and synthetic data. The good thing about human data is it's really high quality. The bad thing is it's not very easy to scale, right? Human cycles are scarce. With synthetic data, the challenges, sometimes it can be hard to teach the teacher, right? If the model is generating data, you need something that's more advanced than the model to move it forward. So pure synthetic data doesn't work. At Turing, humans generate data, but AI verifies it.

>> Got it. - And that's been very helpful.

Jonathan Siddharth

>> And that's going to help agents how?

>> Just being smarter about what the tasks are,

>> task completion? Is that kind of where it goes?

Jonathan Siddharth

>> That's right. I think agents are going to become a lot smarter. Today, at Turing, we train enterprise agents to reach super intelligence. By that, I mean these are agents that can execute complex multi- step workflows in a real world business setting. We've created what we call RL gyms, which is a realistic business environment.

>> That's where they work out, right?

>> Yes. - The agents get their workout in.

Jonathan Siddharth

>> That's right. - They hit the gym.

Jonathan Siddharth

>> That's right. - We all know that.

Jonathan Siddharth

>> Agents train just like humans do, right?

>> Yeah. Yeah. - And in these gyms,

Jonathan Siddharth

>> you have human experts create prompts and create verifiers.

>> And the agents

>> then figure out a trajectory of

Jonathan Siddharth

>> what should be applications I call,

>> what is the sequence in which I should call them to actually get the tasks done.

>> I mean, I could see OpenAI and Anthropics

>> leaning into the big models. For sure, good, good check there. Enterprise, not as savvy. How are you seeing that enterprise market right now?

Jonathan Siddharth

>> So in the enterprise, often, you need smaller models. And often, the enterprise models leverage proprietary data that's inside the enterprise, and proprietary knowledge that's trapped in the minds of humans that are working at the enterprise.

>> Got it. - So in the same way that --

Jonathan Siddharth

>> Like the workflows, knowing the domain

>> expertise, that kind of thing?

Jonathan Siddharth

>> That's right. That's right. And for some enterprise workflows, you may not need a trillion parameter world model. Actually, a smaller model might actually work better.

>> Is the strategy to distill from the bigger models to get there or they bring their own models to the table, or both?

Jonathan Siddharth

>> So oftentimes, we are fine-tuning one of the proprietary or open-source models for that particular use case, and one of the value adds we bring is identifying what's the right model for the problem. For example, John, if you are running an insurance company, and there is a large insurance company that we are working with, this company needed help to automate underwriting and claims processing. So for that, we took a proprietary model. We built an agentic human-in-the-loop system, which is a self-improving system where the agent already automates a lot of the underwriting workflow. But whenever the agent is wrong and the human is correcting the agent, that's becomes training data to improve the next iteration of the agent. So the agent is always improving.

>> Self-reinforcing on that data set.

>> Yes. - Okay. Talk about the business.

>> Where are you guys at right now?

Jonathan Siddharth

>> Revenue wise, can you share numbers, or is that confidential? Or, just give us a sense of your momentum.

>> Yeah. So we are growing super fast.

Jonathan Siddharth

>> We're doing a couple of hundred million in revenues. We are profitable.

>> Nice. Congratulations.

Jonathan Siddharth

>> Thank you. Yeah, it's been great.

>> And after the Scale-Meta news, business has been crazy.

>> Of course. - I don't think I've

Jonathan Siddharth

>> slept in the last few weeks.

>> Well, thanks for staying up

>> all night for theCUBE interview. That Scale AI thing was very interesting because that was basically an acquisition. And so it's just the business has been being distributed around. What does that tell us about the market? Does that tell us that the switching costs are lower, that there's other people that have better products? What does the Scale AI impact do to the market, in your opinion?

Jonathan Siddharth

>> I think there are a few takeaways here, John. First, I think it underscores the importance of data in the raise to AGI, and it benchmarks the value of a lot of these data generation companies. So full credit to Alex and the Scale team for getting here. Second, all these labs, we are seeing for many of our clients, they want to move away and work with a more neutral data partner like Turing. So Turing is Switzerland, where we will work with all the frontier labs.

>> So you need trust. They want trust.

>> They want trust. And a certain level of impartiality,

Jonathan Siddharth

>> that we would prioritize everyone's roadmap equally. So we've benefited hugely from that, where there's a ton of demand that has come in after the news, and we are scaling like crazy to meet the demand.

>> Congratulations. Where are you guys located?

Jonathan Siddharth

>> In Palo Alto. And we have offices in New York,

>> Brazil, Denver. We're opening an office near King's Cross in London.

>> Nice. Well, we're already at our Palo Alto studio there, so we have to have you down for one of our AI leaders events in the future. We do a bunch of them a year, so definitely want to have you in there.

Jonathan Siddharth

>> That's - As customers are building, I want

>> to get your thoughts on your recommendation, because as the transformation happens, skills are being learned. What is the best practice that companies should think about right now in terms of, as you see what you're working on a daily basis, what should they think? They're making generational decisions, how their databases are architected, how their platforms, what they invest in suppliers, whether it's cloud or on-prem, or neo cloud. They're going to get more horsepower. There's more GPU action happening. So yeah, more horsepower is coming. How should they think about setting the table for their future?

Jonathan Siddharth

>> So I'm going to be a bit controversial, John.

>> Yeah. Yeah, bring it on. - So this may not be a popular

Jonathan Siddharth

>> point of view, but this is what I believe.

>> I think CEOs of these companies have to embrace the reality that everything is going to be automated three years from now, everything. When I say everything, I mean every type of work you are doing on a computer, where you are looking at the computer screen, analyzing data, thinking about stuff, using tools to do some tasks, it's going to be automated. My job will be automated completely, right? And to do that, the models have to get better at four things. And those four things, John, are super important. First, multimodality. The models have to understand images, video, et cetera. Second, reasoning. The models have to be able to think and plan and execute a sequence of steps to achieve a goal. Third, tool use. The models have to learn how to use tools just like humans do. And fourth, coding. When the models learn how to code... And then Turing as the company that taught all the LLMs to code, when the models can code, they can create their own tools.

>> Yeah. So we're going to be out of business in three years, so I hope I'm here.

Jonathan Siddharth

>> No, I think... And this is the silver lining, and it's a big one, right? In a world where every human is 100x more productive, every human can do 100x more things than they do today. For example, when I am maximally automated today, I have the bandwidth to run one company.

>> Okay. - I might be running a hundred companies.

>> Right now, we're live streaming this video.

Jonathan Siddharth

>> Yeah. - We can ingest the linguistics.

Jonathan Siddharth

>> Yeah. - We can call some agents that could wrap

>> around the screen, kind of like a Bloomberg terminal --

Jonathan Siddharth

>> Yes. - ...

>> and bring in other data.

>> As you speak, instead of saying captions,-

Jonathan Siddharth

>> Yeah. - ...

>> it's got discovery data.

Jonathan Siddharth

>> Yeah. - Bookmark that.

>> So you have this new user experience-

Jonathan Siddharth

>> Yeah. - ...

>> coming around the day. That's what you're getting at.

Jonathan Siddharth

>> So my job may go away on a virtual basis, yeah,

>> cutting up some videos, packaging content. I see that. No problem.

Jonathan Siddharth

>> Yeah. - I was just talking to an entrepreneur out

>> there about this one point. But my role of getting original, trusted content-

Jonathan Siddharth

>> Yeah. - ...

>> is still valuable, to your point.

Jonathan Siddharth

>> It is still valuable, and you sound just like a Turing customer. Your vision for how you would automate your job, that's exactly what our customers do that.

>> Okay. I'm going to be a customer. We'll do a trade.

Jonathan Siddharth

>> That's great. - We're self-funded. We're not very

>> a lot of cash,

>> so yeah. Yeah, but this is where it's going.

Jonathan Siddharth

>> Yes, that's where it's going. And also, John, you'll be able to scale yourself.

>> Yeah. - Today, you're limited by, okay,

Jonathan Siddharth

>> we are in Paris here right now.

>> There's only one John, and you're in this place, but there's a future where digital twins or digital surrogates of you are doing interviews. There's a version of you in the --

>> Yeah. - --

Jonathan Siddharth

>> Palo Alto office, completely virtual,-

>> Yeah.

Jonathan Siddharth

>> Yeah. - -- that's maybe interviewing Dario from Anthropic.

>> Yeah. Yeah. We're definitely on the same page on that one.

>> I believe that we're going to have a physical- digital convergence.

Jonathan Siddharth

>> Yeah. - And that is first party.

Jonathan Siddharth

>> Yeah. - And I say first party, I mean like

>> as we treat relationships-

Jonathan Siddharth

>> Yeah. - ...

>> but with context.

Jonathan Siddharth

>> Yes. - The behavior shifts to avatar images.

>> I mean, I have a bot working right now in our lab

>> that's basically a human host that looks like a human. It's not human, but it could do intro packages. It can go call a script and say, write me a script. It knows the interview pay clip maybe and does a payload. Here's a script. "In this next segment, John is going to talk about the four things. It's controversial. Stay tuned." I mean, that could happen-

Jonathan Siddharth

>> Yeah. - ... right now.

Jonathan Siddharth

>> Absolutely. - Okay, so as customers' minds get blown,

>> how do you rank your customers right now?

>> Because you have pioneers, you've got fast followers, and you've got the settlers who come in after. It's a classic business school definition of kind of progression. How would you rank your engagements right now? Because I'd see more growth certainly coming for what you're doing. What percentage of your customers fall into the pioneer category, fast followers, and then the settlers who come on and get settled, go mainstream?

Jonathan Siddharth

>> Yeah. So we have two categories of customers. On one hand, we work with all the model builders.

>> They're the pioneers. - They're the pioneers.

Jonathan Siddharth

>> On the other hand, we work with Fortune 500 clients in banking and financial services, retail, pharma, these traditional industries that solve really important problems, but need a guide to help them navigate the AI transformation.

>> Yeah. That's a fast follower category by their appetite.

Jonathan Siddharth

>> Yeah. - For instance, financial services has always been

>> a fast follower, early adopter.

Jonathan Siddharth

>> Correct. Correct. - Life sciences, automotive,

>> they love digital twins and super computing-

Jonathan Siddharth

>> Yeah. - ...

>> because they never had it before.

Jonathan Siddharth

>> Yeah. - So for the first time,

>> they have democratized super computing-

Jonathan Siddharth

>> Yeah. - ...

>> as Jensen Wong always talks about.

Jonathan Siddharth

>> Yeah. - So I see that as kind

>> of like, well, you agree with that.

Jonathan Siddharth

>> Yes. - And then you have the rest of,

>> I have an IT department,-

Jonathan Siddharth

>> Yeah. - ...

>> that's just transformation.

Jonathan Siddharth

>> Yes. - That's kind the progression. You agree with that?

Jonathan Siddharth

>> That's right. That's right. And I think the AI

>> transformation is going

>> to look very different from the cloud transformation and the mobile transformation. Cloud and mobile were a lot more about doing the same thing that you were doing, but with more efficiency.

>> Yeah, same as the web on an app-

Jonathan Siddharth

>> Yeah. - ... in an app store.

Jonathan Siddharth

>> Yeah. - Here's enterprise from day one

>> because you have a data problem.

Jonathan Siddharth

>> Yes. - Data feeds the AI.

Jonathan Siddharth

>> Yes. - That's where you guys come in.

Jonathan Siddharth

>> That is correct. Data feeds AI.

>> And in enterprises, oftentimes the biggest bottleneck

>> to building a generative AI application is

>> that the data is not in a usable form.

>> Either the data is siloed or the data is not being logged. So we meet our clients where they are and help them, first, build the right data infrastructure to build these fancy agentic systems on top.

>> Johnathan, great pleasure. We definitely will do a deep dive with you back in the studio in Palo Alto, or wherever you are when you're not serving your massive growing customer base. Congratulations on the success. Thanks for coming in and sharing your mixture of expert perspective here on theCUBE. Thank you for coming on.

>> Great. - Appreciate it.

Jonathan Siddharth

>> Thank you, John. Thanks for having me.

>> Okay. I'm John Furrier with theCUBE.

>> We're bringing all the action, doing our part,

Jonathan Siddharth

>> sharing the data with you here on theCUBE, here from Paris, live, thecube. net. Thanks for watching.