Joe Kim, CEO of Office Hours, on the end of crowdwork


Background
The demand from frontier AI labs for expert data labeling has driven the growth of companies like Mercor ($100M annualized revenue in February 2025) to Invisible ($134M in 2024) to Handshake ($50M from AI data labeling).
To learn more, we chatted with Joe Kim, CEO & co-founder of the expert marketplace Office Hours ($5M seed round, CRV).
Key points via Sacra AI:
- Traditional boundaries between expert calls (GLG, AlphaSights, Tegus), market surveys (Gartner, Forrester) and user interviews (UserTesting, User Interviews) are collapsing into a single "primary research" category worth tens of billions—with life sciences consulting alone at $50B+—as AI abstracts away specific formats into agents that can run voice interviews & surveys, generate & extract content & data and produce completions & reports. "A lot of this stuff is collapsing in real time in terms of TAM spend and inverting in different ways. . . Just look at the B2B surveys space and market research space. This is blending right now. Calls, surveys, questionnaires, interviews, conferences—these are all just different form factors for the same thing, which is humans learning from each other. The total primary market research space is in the tens of billions of dollars.”
- As foundation model labs have moved towards reasoning models which require PhD-level evaluation and fine-tuning, aggregating the supply side of verified & highly-credentialed domain experts across finance, legal, healthcare & other verticals has become key to serving demand—versus low-cost annotations from crowdsourced Mechanical Turk-style overseas workers. "At a niche level, whether you're talking about banking or life sciences or legal, the existing data labeling players in the market today are all based on crowd work. . . But [now] there is more and more demand for quality assurances around expert quality and credentialing. It’s all moving very fast, and the form factor has to be different from the traditional form factor for data labeling. . . you need to offer some level of education and credentialing and expertise.”
- Just as LinkedIn built recruiting, sales, and ads on the "professional graph of what people do," expert marketplaces like Office Hours & Intro, AI recruiting marketplaces like Mercor, and job markets like Handshake are converging on the "professional graph of what people know," integrating expertise with monetization for a platform on which applications can be built for B2B user research, recruiting, mentorship & training, and AI model training. “The end-all mission is to help knowledge professionals monetize their unique insights. . . Using the LinkedIn analogy, they built a billion-dollar business based on that economic graph—a billion-dollar sales business, recruiting business, ads business, and social network. There are multiple different use cases that you can expand the TAM on when you build a true open expert network.”
Questions
- I'll start with the origin story. You were at AlphaSights for a while. What were the experiences and insights you had that led you to start Office Hours?
- You spoke to a couple of problems to solve, like the incentives problem. What would you say is the biggest nut to crack where if you crack that nut, everything else is easy? What is the biggest bottleneck in terms of efficiency in your self-serve model?
- Is it the case where, as in a lot of marketplaces, the supply side is tougher than the demand side or the inverse?
- If we could zoom out, how would you lay out the landscape of the market for experts and their insight? Who are the players? How would you taxonomize them, and where does Office Hours fit in?
- How do you size the TAM for this broader expertise graph?
- In terms of the use cases, you mentioned mentorship and B2B interviews of potential buyers. Are there other use cases that you've seen growing quickly in the last twelve months?
- How does that AI interviewing tie in with Office Hours? What are your views about how this modality will spread across expert calls and leveraging human expertise?
- So you don't plan to introduce the ability for the demand side to spin up AI agents to interview folks on Office Hours?
- In terms of the network effects, how would you talk me through the network effects specific to your model and how you see them accelerating with scale?
- Do you see the knowledge locked in people's heads becoming more valuable as LLMs commoditize public knowledge? What are the challenges in extracting that value? Is it variability in quality, or is it something else?
- Office Hours has not yet launched a transcript library. What do you think about transcript libraries in terms of enhancing the value of the product as a whole and the offering as a whole?
- Touching on AlphaSense, they have financial models, earnings reports, broker reports, financial data, and then on top of that—they layer in the transcript library and on-demand expert interviews. Someone might look at that model and say, "This is the future. Expert interviews are a feature, not a product." How would you argue with someone that takes that view looking at AlphaSense?
- What metrics would you watch to gauge whether Office Hour’s expert network model is gaining traction. What would be the top metrics that you might look at to see that the model is working?
- Earlier you mentioned new use case expansion. Recruiting companies like Handshake and Mercor have made inroads into this LLM post-training space. Is there a market there for expert networks like yours that are building on human insight?
Interview
I'll start with the origin story. You were at AlphaSights for a while. What were the experiences and insights you had that led you to start Office Hours?
I was a super early employee at AlphaSights. Back then, GLG was the main 800-pound gorilla in the industry. We came on the scene, and I started in New York. We were largely servicing hedge fund clients on Wall Street, so I wore a suit and tie every day.
The product we sold was to hedge funds saying, "You're thinking about taking a multimillion-dollar position in something relatively soon. You have this perspective on something. Wouldn't it be helpful if we could somehow put you in touch with experts, boots on the ground, who actually touch and feel the thing that you're potentially going to allocate capital into or take a position on? We can get them on the phone for you tomorrow to pick their brain." Intuitively, the answer was—absolutely. This product had incredible product-market fit.
When we started, we began teaching private equity as an asset class how to use this product as well. Private equity as an asset class grew tremendously over the next ten to fifteen years, and we rode that wave. Private equity taught consulting firms, so the Bains, McKinseys, and BCGs of the world had to do expert interviews as well to acquire hard-to-reach insights to help with diligence and decision-making.
The actual product and value proposition we were providing to our customers was incredible and profound. If I have a question—it could be a very important question—there are perspectives all around the world that could help me make a decision around this question.
And there's a market-clearing price that I would be willing to pay to answer that question or have some perspective on it. It's really just a search, match, discovery, identification problem. It's a connectivity problem and an incentivization problem. Those are the frictions involved with creating this market. But if that's solvable, that's a pretty compelling, intuitive way to help you make decisions.
Over time, I saw that it's not just hedge funds or private equity firms with important questions. All people have important questions, even in the professional world—operators, founders, executives, and entrepreneurs. There's demand for this product across multiple different use cases.
Then I saw a pain point as well. I had a bunch of my consulting clients over the years who would eventually leave consulting, go to business school, and then after that join an early-stage startup, start a company of their own, or start a fund. These are folks who were trained in the behavior of using expert interviews to make business decisions. There are about 40,000 of these ex-consultants out there in the world today since expert network interviews as a product was created.
But the model for AlphaSights was always a professional services model, so you want to work with a high concentration of high-spend customers and really build client relationships with them. That's how you grow.
These ex-consultants—they may not be a natural customer for AlphaSights, but they still need this product. It's an incredible, profound thing to help them make decisions. The pain point they had was that they tried to solve it themselves by using LinkedIn, using friends, reaching out to potential experts they would like to speak to, trying to figure out who is the right person to reach out to, explaining what they're trying to do, asking them for time, and trying to incentivize them by offering gift cards or taking them out to lunch. It's a very fragmented, broken, and friction-filled process. Solving that pain point lends itself well to the magic of online marketplaces—it reduces friction and increases trust.
The final piece is a dynamic I saw when I first started at AlphaSights: A lot of the job was convincing subject matter experts that they were experts themselves and that someone was actually willing to pay them to talk to them, and that this wasn't a scam—that you could actually earn income outside of your nine-to-five job using the Internet economy and gig economy. Over time, this perception has evolved and it has become much more desirable for knowledge workers to figure out ways that their unique value can help other people, and potentially create value from that.
That side lends itself well to another opportunity, which is using technology to allow the supply side to be more self-service. People are looking for ways and channels—they're doing YouTube videos, Substacks, and TikToks to try to teach other people the unique perspective that they have. Enabling that gave us the opportunity and insight to build a tech-enabled platform.
You spoke to a couple of problems to solve, like the incentives problem. What would you say is the biggest nut to crack where if you crack that nut, everything else is easy? What is the biggest bottleneck in terms of efficiency in your self-serve model?
To take the converse of that question, the biggest operational bottlenecks for the traditional expert network space is that they are professional services firms. Everything is about the people—their employees. Like a law firm or accounting firm, it's all about the talent and scaling talent in a professional services-way. Everything is determined by talent density. These are professional services firms, and their client service talent is what drives success. Technical talent is almost like a second-class citizen—it's all in service of assisting professional services.
For us, we always had this thought that this is a product-driven business that can be scaled by product. Everything we do is from a product lens. We're going deep on user journeys on the customer side and the expert side. Everything we're doing is to increase trust and reduce friction, period.
Professional services is one way to do that. You can trust another human being who's wearing a suit and tie for you, who sounds intelligent and is proposing these experts to you. That's the trust layer. There are compliance components to it as well—quality, screening, matching, all those different things.
Then there's friction on both sides. How easy is it to find this person? How easy is it for me to share what I know and explain to someone what I do? How easy is it for me to schedule an opportunity or do a call? How easy is it for me to browse different opportunities?
Every single thing that we do, every person that we hire, is tasked to either reduce friction or increase trust. We aim to scale that broadly and infinitely over time with our platform.
Is it the case where, as in a lot of marketplaces, the supply side is tougher than the demand side or the inverse?
Not necessarily. This goes into what people misunderstand about the expert network industry. Whether it's Wall Street or even the incumbents, this can be much more of a technology-enabled business than the standard model bears today. It's highly scalable. What we're trying to do is solve a search, match, discovery, connection, and incentivization problem. If you solve it, you can unlock this otherwise underutilized asset in the supply side. Technology is the big driver here.
There are a lot of preconceived notions that you need a human in the middle to be able to build trust between two parties in brokering what we sell, which is human insight. We're seeing in real time that that's not the case.
If we could zoom out, how would you lay out the landscape of the market for experts and their insight? Who are the players? How would you taxonomize them, and where does Office Hours fit in?
Traditional players serve traditional customer types—folks that serve private equity firms, hedge funds, and management consulting firms. I would put GLG and AlphaSights into those buckets, but there are a few other big ones: ThirdBridge and GuidePoint, for example. They're pretty sizable, with hundreds of millions of dollars in annual revenue.
Then there's Tegus, which started in their own category where they weren't selling experts—they were selling transcripts, a library of transcripts. The way they were getting those transcripts was in a genius way by running an expert network, but really they were a transcript library. With the acquisition by AlphaSense, I would bucket them away from a traditional expert network and more in their own category. From my understanding, AlphaSense is a primary research tool specifically for financial services. It's a large TAM, but the expert network portion becomes smaller and smaller in terms of their vision.
Then there's Office Hours, which is squarely in a different category. Really what we're doing is offering a platform that offers access to human reasoning. Our mission is to help the billion knowledge professionals in the world realize the true value of what they know.
We definitely compete with traditional expert networks because our network is highly capable of producing the high-quality expertise needed to service a hedge fund or consulting firm. But the end-all mission is to help knowledge professionals monetize their unique insights. This looks like multiple different use cases, including diligence, but one of our biggest use cases for our customers is actually B2B user insights. We have a lot of different types of B2B software companies—early stage, late stage, public companies—that use Office Hours to access target buyers, learn from them about their pain points and incumbent vendors, and test out new features and marketing language.
If you zoom out, the same way LinkedIn built the graph of what people do where I can search for you by the company you work for or your title, we're building the graph of what people know. Our fundamental belief is that there's a lot of economic value that can be created by searching, matching, identifying, and connecting with people based on the things that they know about.
Using the LinkedIn analogy, they built a billion-dollar business based on that economic graph—a billion-dollar sales business, recruiting business, ads business, and social network. There are multiple different use cases that you can expand the TAM on when you build a true open expert network. In that category, there's a company called Clarity.fm and a company called Intro that are taking different angles to making it easy to access human knowledge on demand.
How do you size the TAM for this broader expertise graph?
A lot of this stuff is collapsing in real time in terms of TAM spend and inverting in different ways. AI plays a big role in this as well. The traditional expert network space is about $3 billion in spend and growing. We think building a highly network-effects liquid, better user experience, software-enabled, tech-enabled, AI-enabled, mobile-first platform that reduces friction and increases trust will allow us to service that market very well.
But there are also adjacencies. Just look at the B2B surveys space and market research space. This is blending right now. Calls, surveys, questionnaires, interviews, conferences—these are all just different form factors for the same thing, which is humans learning from each other. The total primary market research space is in the tens of billions of dollars.
Then you could also look at life sciences consulting alone. The consulting services for life sciences is $50 billion plus. Those are all areas where learning from people has direct value, and these are all just the form factors that have been put in place because that's just the way the market has solved this problem: how do we get the insights from the stuff that's stuck in people's heads? You can package that insight into a report via Sacra, for example—there are all these different form factors to do that.
The promise of the Internet was always to connect people directly to each other for the benefit of society. We think technology has a really big role in being able to flatten a lot of the stuff that makes direct access much easier as well as referring a lot of that newly created value to the experts themselves.
If you have a graph of what people know and can connect easily to that and trust that that's true in the same way I trust that if you put something on your LinkedIn is true, there comes a certain point where people want to hire contractors that way. They will want to hire based on what you know and your experiences and your unique insights, versus where you worked. Recruiters are interested in hiring people full-time that way. It's a great way to source your next talent pool based on the things that they know about.
We also have a bunch of customers who are using this for mentorship. They're enterprise companies who buy packages for their employees, and their employees can book expert sessions once a quarter or something like that. This is not just a generalized expert, say, in marketing: we have enough liquidity where you can find someone very specific and niche to your situation. Maybe you're an engineering manager that just got promoted, struggling with specific problems you're trying to solve—you can find folks who have actually solved those problems in the past.
Finally boiling it down further, there’s a human insights market in the AI space as well. There's a lot that human knowledge, niche and deep subject matter expertise, intuition, wisdom, judgment, and reasoning can do that can help accelerate AI's ability to reason like humans as well.
If we can really play a part in supporting creating new value from human knowledge across all these areas—it's almost like there are infinitely fractal areas that we can move into and support.
In terms of the use cases, you mentioned mentorship and B2B interviews of potential buyers. Are there other use cases that you've seen growing quickly in the last twelve months?
A lot of user research. A lot of B2B user research. This is where AI interviews, if you've heard of this concept, are really taking the market by storm.
The old school way, a year ago, the traditional way of doing B2B user research is you would create this complex survey that you could send to a trusted cohort of experts or users or potential users, whatever your demographic is, to answer questions that could help you make product-level and pricing-level decisions as well as marketing decisions or simply gain competitive intelligence.
What generative AI has allowed us to do, this voice-to-AI technology, is get more nuance from those surveys. The form factor looks less like just quantitative multiple choice answers or structured answers, but it can actually have a dialogue with somebody because it can double-click into a specific answer that someone might have. It can gear the audience in the right direction in terms of questioning. It can have much more complex reasoning in terms of conditional questions to ask or follow-up questions to ask.
What that is doing is increasing TAM for human insight like crazy. All of a sudden, user research teams don't have to spend forty or fifty hours on expert interview calls to talk to each person individually. They can conduct all of these interviews using AI but have confidence that they'll get nuanced answers and color, which is really what they're looking for. On top of that, it's much easier to synthesize—you don't even have to read the transcripts anymore. You can have AI synthesize them.
This is another technological tailwind that creates more value in human knowledge because it's increasing the trust in the results that you're getting, but it also reduces the friction to be able to go and activate research on demand.
How does that AI interviewing tie in with Office Hours? What are your views about how this modality will spread across expert calls and leveraging human expertise?
First, it's a TAM expander, it's a tailwind for human insight and general access to human insight. Number two, it's a commodity already. There are voice-to-AI APIs that you can use. It's incredible technology, and it's getting easier and easier every single month to implement.
There's a perspective that depending on how much data you have in your proprietary transcript libraries, the AI interviewer will somehow be superior to other AI interviewers that you can just buy off the shelf via an API. There are some questions around that. There's a chance that an AI backed by a new generalized model will conduct interviews as well as a highly trained hedge fund analyst AI interviewer or AI agent.
But at the end of the day, the value of doing an expert call is about the types of questions you ask and the style of questions you ask. So we think it is probably more likely that customers are going to prefer to build their own AI models for this use case. A big hedge fund would rather train their own AI interviewer on their analysts' questions and answers versus buying someone else's opinion on how to do that.
The last piece is that experts really appreciate this form factor. It's much more nuanced than a survey. It's either voice or interviewees can type and chat back and forth with an AI interviewer. Either way the interview becomes asynchronous, which is incredibly important. It's a subtle point, but scheduling is one of the main frictions that experts have: I have to turn on my video, I have to find time out of my day, a sixty-minute block where I'm available. With an AI, you can start it, pause it, go pick up your kids, then start it again. That form factor is pretty incredible in terms of reducing friction.
It's great and easy to do, and it will be done. But it's not just going to be easy for the networks themselves to do. It's going to be easy enough for customers to do themselves and to have their own AI interviewers that they prefer.
So you don't plan to introduce the ability for the demand side to spin up AI agents to interview folks on Office Hours?
We do that already. But it's going to be commoditized. Every expert network in the next six months will do this because it's not hard to do.
In terms of the network effects, how would you talk me through the network effects specific to your model and how you see them accelerating with scale?
It's pretty intuitive. Classically, the more customers there are, the more valuable it is for experts. The more experts there are, the more valuable for customers.
Our double-click into that is the way we think about building our business. The atomic unit on the expert supply side is human insights, and on the demand side it's human questions. It's less about user network effects—that's a great proxy or KPI—but it's also the insight network effect. That's the atomic unit we're going for. How many insights can we provide at any given time, and how many questions can we provide a market for those insights at any time?
So there are more network effects than just the users themselves, it’s also in what the users provide.
On same-side network effects, the most powerful differentiator we have is that we are the only open expert network in the world. I'm a physician, and before I sign up to Office Hours, I can search on Office Hours. I'm an oncologist, and I can search for other oncologists. I can see folks I met in med school or from residency or fellowship or my previous job or I met them at a conference. I can see how many Office Hours they've done, I can see their profile, I can see testimonials that they have. I can see what their rates are and stuff like that. That creates increased trust.
That's a core differentiator, which allows us to lean into deeper network effects than the traditional black box brokerage expert network.
Do you see the knowledge locked in people's heads becoming more valuable as LLMs commoditize public knowledge? What are the challenges in extracting that value? Is it variability in quality, or is it something else?
The variability comes from two things. It's the context in which you are trying to learn about something, which dictates the value of that human expertise. From an investment perspective, if you're investing in public school educational software, then all of a sudden, superintendents of a school district are incredibly valuable to you in that context—their perspective and insight of what products they like to use and how they think about buying those things. The next day when that deal is done, the value of that knowledge contextually maybe decreases over time in terms of premium.
The second thing is the way you ask questions is incredibly important. To do that, you need to know what you're looking to achieve from that conversation.
We have a motto at Office Hours: everyone's an expert, they just don't know it yet, and our job is to unlock that value. But it's for both sides. It's to help experts truly realize, "Here, you know these things. These are unique to you. The only reason you know them is because of the unique experiences that you've accrued over time." It's also to help our customers really solve for what questions they should be asking. Part of the idea is turning unknown unknowns to known unknowns. Then, once you know what you don't know, here are the resources at your fingertips to be able to learn about those topics.
Every single customer has access to ChatGPT already and they're already using LLMs to learn a ton, and they should be. But they're still using Office Hours because there’s a need to even know what questions to ask.
On the expert side, this dynamic gets at the core of what you are an expert on—it's your taste. No one knows your taste better than you, your opinion, your judgment, your wisdom, your intuition. These are the types of things that get at the core of unique human insight, which continues to be valuable. The way that you reason continues to be valuable in specific contexts.
Warren Buffett has a great quote. He was interested in the coal mining industry, and his team was putting together all the financial models, gathering all the data, gathering all the information, putting that together, creating insights from that. But he also sets up one-on-one conversations, meetings face-to-face with the CEOs of the top 10 coal mining companies. He asks each of them a really cool question: "If you could invest long on a company outside of yourself, one of your competitors, which one would it be? If you could short one, which one would it be?" He asks each individual CEO the same question.
That's a great example of asking a specific question, knowing what you want to get out of that answer, and understanding that that's just accrual of unique insights and perspectives and inside information and gut and instinct that you're collecting. Each CEO has their own unique insight, but it's valuable collectively just the same.
As AI becomes more powerful, there's going to be less and less information that humans need to give you, but they're always going to have their unique experiences, and those things become valuable based on the context in which you want that knowledge.
Office Hours has not yet launched a transcript library. What do you think about transcript libraries in terms of enhancing the value of the product as a whole and the offering as a whole?
Transcript libraries are very valuable because they reduce friction to insight. We do not currently offer a public transcript library. We have customers who have thousands of their own transcripts, which they can access and get insights from at any time. That's actually really helpful when we have customers who have thousands of employees that do expert calls. To be able to see the transcripts across all those different teams actually becomes really valuable for their own personal library.
Our mission is to help everyone realize the true value of what they know. We would like to build the ability for experts to be able to earn money based on transcript value, whether that looks like a royalty or views or even just demand generation for future calls. That's the system that we're trying to build. Without that, we become a transcript library business where the focus is less on building the core user experience for the supply side. Our mission is building the best expert network for experts possible.
There's definitely a way to create a transcript library where that's the case, but we want to make sure that we're aligned on value. The value of the transcript library is only as good as the experts themselves and the conversations they can have. That's what generates the transcripts themselves. It's a really hard business already to just build a really high, liquid, robust, high-quality, highly engaged network of high-quality experts. Our focus is there.
There's a world where AlphaSense is a customer of Office Hours. When they need transcripts, or they need experts, where can they get experts from? From Office Hours.
I don't know if transcripts are the right form factor. Transcripts are really helpful to get really smart on something, but ChatGPT is getting really good at getting smart at things as well. Treating each transcript individually, you get a lot of good insight from it. But everyone's just talking, and it's not facts that you're getting. There’s rounding errors and color and people's opinion on things. I don't know what that looks like when you aggregate that. You're not going to be able to model off of that really accurately, for example.
Touching on AlphaSense, they have financial models, earnings reports, broker reports, financial data, and then on top of that—they layer in the transcript library and on-demand expert interviews. Someone might look at that model and say, "This is the future. Expert interviews are a feature, not a product." How would you argue with someone that takes that view looking at AlphaSense?
Looking at it from AlphaSense and thinking about who their buyer is, AlphaSense is very much a financial services tool set, trying to be the most comprehensive tool set. Within that tool set you also have primary research expert interviews, and that's really great.
We try to take a more holistic approach where the job that we are hired to do is in service of our experts. What we're trying to do is give them as many opportunities to share what they know and create value from it as possible. The expert network industry space is a great space for people to earn. There's a premium on that type of value, that type of knowledge, and that human reasoning.
We take a more holistic view in terms of what we can offer our experts and our talent pool—the ability to earn, to share what they know, and benefit others by sharing what they know. That is all within our remit. I would maybe agree that a traditional expert call is one of the ways we offer value to them—it’s a product that we offer our experts, and we consider it a feature to them as well. Here's one of the ways that you can earn income: talk to a hedge fund and share what you know about your experiences in the industry. But it won't be the only way for us. We’re focused more broadly on how to create economic opportunity for the supply side.
What metrics would you watch to gauge whether Office Hour’s expert network model is gaining traction. What would be the top metrics that you might look at to see that the model is working?
Retention is a huge one, and that's actually one of the core costs of a traditional expert network—it's pretty low retention for high-quality experts. Typically, traditional expert networks are spending a lot of time trying to backfill or add net-new experts to their expert network. A big reason why is because folks just end up having a bad experience after one or two expert calls. They come out thinking: There's no amount of money where I would do that again. There's just so much friction involved in the entire process.
The number one thing for measuring supply side happiness is retention. In order to drive retention, it's the user experience. How do we reduce as much friction as possible? Then it's also tracking opportunities. How many opportunities at any given time are accessible to our experts in terms of earnings?
Our north star is the number of knowledge-sharing interactions, whether they're one-on-one phone conversations, an AI interview, a mentorship call, a user interview, or a survey.
Earlier you mentioned new use case expansion. Recruiting companies like Handshake and Mercor have made inroads into this LLM post-training space. Is there a market there for expert networks like yours that are building on human insight?
Things are moving really fast. The models are getting more sophisticated than ever. They're demanding higher levels of sophistication in terms of fine-tuning and evaluations. We've moved past the crowd-work style of data labeling and annotation to improve model effectiveness. At the same time, the expectation from the world is that these AI models will be able to do more functional-level and expert-level tasks.
At a niche level, whether you're talking about banking or life sciences or legal, the existing data labeling players in the market today are all based on crowd work. That's their model. That's how they built their businesses around that. But there is more and more demand for quality assurances around expert quality and credentialing. That becomes a pain point. But it's the same pain point that we've been talking about since the very beginning, which is access to hard-to-reach human insight. The ways that we do that on Office Hours is through solving search, match, identification problems, connectivity problems, incentivization problems The way that we can solve these problems is to increase trust and reduce friction.
We are well-placed, and our network is well-placed to be able to provide all sorts of human insight for those specific uses. It’s all moving very fast, and the form factor has to be different from the traditional form factor for data labeling. That's why Handshake has done so well. They have access to PhD students. Because these models are getting so sophisticated, you need to offer some level of education and credentialing and expertise.
There are all these different types of reinforcement learning platforms that are starting to be built as well that will allow higher levels of sophistication and different types of training and fine-tuning that these foundation models are demanding now. You can just project forward in terms of what's needed, all the form factors that might emerge, and the new opportunities for providing hard-to-reach human insight that will be created.
Disclaimers
This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.