Jul 24, 2025 · 32m · no-priors
No Priors Ep. 124 | With SurgeAI Founder and CEO Edwin Chen
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of No Priors, Surge AI founder and CEO Edwin Chen discusses how his bootstrapped data company achieved over one billion dollars in revenue by supplying critical human data to frontier AI laboratories. Chen shares key insights on scalable oversight, the limitations of synthetic data, the flaws of benchmark hacking, and the future of specialized foundation models.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 24.4% of the talking time here. How this is scored →
The hosts as informed peer 4.1 Guest teaching 4.6 Guest disagreement 3.2 The hosts pushing back 2.1
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:48 Exposing metric gaming in frontier labs
Edwin aggressively calls out frontier lab leadership and researchers for intentionally trading model factuality for superficial leaderboard gains on LMSYS.
Hardest push from the hosts ▶ 4:27 Elad counters Edwin's anti-fundraising stanceElad directly pushes back on Edwin's broad rejection of venture capital by highlighting that founders outside Silicon Valley suffer from severe under-capitalization.
Biggest teaching moment ▶ 29:38 Deconstructing naive definitions of data qualityEdwin breaks down why standard industry practices like checkbox counting and hiring PhDs produce terrible training data, reframing quality from first principles.
The host holds their own ▶ 20:47 Elad analogizes model training to directed evolutionElad brings technical depth to the discussion by drawing a direct parallel between reinforcement learning objective drift and unexpected artifacts in protein evolution.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Surge's Founding Thesis and Origin Story | 5 | 2 | 4 | 4 | Edwin launches a fierce critique against Silicon Valley founders raising capital for vanity rather than solving problems. Elad pushes back with investor nuance, noting that while SV over-funds, companies outside SV often suffer from under-funding. | |
| Early Team Building and Hiring Mistakes | 5 | 4 | 4 | 4 | Sarah and Elad press Edwin on the practical hiring challenges unknown founders face without venture backing. Edwin strongly dismisses early PM and data science hires, arguing founders should stay hands-on. | |
| Surge's Core Product Offerings and Deliverables | 3 | 4 | 1 | 2 | Sarah intervenes to anchor the conversation on Surge's actual products and revenue breakdown. Edwin systematically explains SFT data, verifiers, preference data, and failure analysis. | |
| Differentiating Surge from Body Shops via Technology | 5 | 5 | 3 | 2 | Edwin disparages legacy data vendors as mere body shops with low quality ceilings. Elad asks a technical question regarding how Surge scales evaluation loops without running out of human evaluators. | |
| Scalable Oversight and Human-AI Collaboration | 4 | 5 | 1 | 1 | Sarah asks about RL environments and scalable oversight interfaces. Edwin provides concrete examples of multi-tool enterprise environments simulating real-world salesperson workflows. | |
| Synthetic Data Limits, Superhuman AI, and Alignment Traps | 7 | 6 | 4 | 2 | Edwin critiques the synthetic data bubble and labels LMSYS arena a plague on AI due to superficial vibe evaluations. Elad demonstrates deep scientific expertise by connecting model objective collapse to protein evolution selecting odd local maxima. | |
| Human Evaluation as the True Gold Standard | 3 | 4 | 3 | 2 | Sarah pushes for realistic alternatives to public benchmarks. Edwin insists rigorous, fact-checked human evaluation remains the only reliable gold standard against clickbait training. | |
| Competitive Landscape and the Meta-Scale Deal | 3 | 4 | 3 | 1 | Sarah asks about the competitive fallout of the Meta-Scale deal and future model specialization. Edwin asserts that low-quality vendors burned customers and predicts specialized frontier models rather than single-commodity AGI. | |
| Benchmark Hacking and Future Public Research | 3 | 6 | 5 | 1 | Edwin reveals that frontier lab researchers deliberately make models worse to climb LMSYS and IFEval leaderboards by inflating response lengths and emoji counts rather than building real capabilities. | |
| Defining High-Quality Data Beyond Checkboxes | 3 | 6 | 4 | 2 | Elad asks Edwin to define true data quality. Edwin rejects checkbox rubrics, Craigslist hires, and English PhD credentials, explaining why first-principles curation is necessary to prevent scaling mediocrity. |
Statements from this episode (19)
Chen: Surge AI Is Largest Human Data Player in AI
“We are kind of like the biggest human data player in this space.”
Chen: Surge AI Operates With Just Over 100 Employees
“And we're about a hundred, a little over a hundred people.”
Chen: Surge AI Was Profitable From Day One, Avoiding Venture Capital
“I think we were very, very lucky to be profitable from the start. So we didn't need the money. It always felt weird to give up control.”
Chen: Many YC Founders Only Care About Fundraising Brags and Headlines
“Like, if you talk to a bunch of YC founders or whoever it is, like, what is their goal? It really is to tell all their friends that they raised ten million dollars and show their parents they got a headline on TechCrunch. Like, that is their goal.”
Chen: Early Startups Should Not Hire Data Scientists as First Employees
“Like I would never hire data scientists when the first three people in a company. And I say that because I used to be a data scientist. Like data scientists are great when you want to optimize your product by two percent or five percent, but that's definitely …”
Chen: Early-stage startups should not hire product managers
“Product managers are great when your company gets big enough. But at the beginning, you should be thinking about yourself about what product you want to build. And your engineer should be hands-on. They should be having great ideas as well.”
Chen: Most AI Data Competitors Are Non-Technical Body Shops
“A lot of other companies in this space, they are essentially just body shops. What they are delivering is not data. They are literally just delivering warm bodies to companies. And so what that means is like at the end of the day, they don't have any technolog…”
Chen: Generative AI Data Has an Almost Unlimited Quality Ceiling
“So there's a very, very low ceiling on the bar of quality, but then take something like writing poetry. Well, I suck at writing poetry. Hemingway's definitely want to write a much better poem than I am. Or imagine, I don't know, a VC pitch deck. You're going t…”
Chen: AI RL Environments Are Too Complex to Create Synthetically
“I think one of the things that people really underestimate is how it is, how complicated it is that you can't just synthetically generate it.”
Chen: Synthetic RL Environments Alone Won't Meet Future Frontier AI Demand
“Like, I don't think our environments alone will suffice just because, I mean, it depends on how you think about our environments, but oftentimes these are very, very rich trajectories are very, very long. And so it's almost like inconceivable that a single rew…”
Chen: Surge AI Heavily Uses Synthetic Data to Supplement Human Labelers
“Like we use it like a ton ourselves in order to supplement what the humans do.”
Chen: LMSYS Chatbot Arena is a giant plague on AI
“One of the things I think is a giant plague on AI is Elimsis, Elimarena.”
Chen: Surge AI Works With All Frontier Labs on Model Evaluation
“So internally we do a lot of work actually today with working with all the frontier labs to help them understand their models. So again, we're constantly evaluating them. We're constantly surfacing loss areas for them to improve on and so on and so on.”
Chen: Meta's Deal With Scale AI Actually Benefited Surge AI
“It's been beneficial because yeah, there were still some legacy teams using scale. Like they just didn't know about us because we were still pretty under the radar.”
Chen Bets on xAI to Catch Up With Top Frontier Labs
“So I would bet on XAI. I think they're just very hungry and mission-oriented in a way that gives them a lot of really unique advantages.”
Chen: Frontier AI Models Will Proliferate Rather Than Commoditize
“So I actually see more and more frontier models opening up over time because I actually don't think that the models will be commodities.”
Chen: Frontier AI Labs Are No Longer Publishing Research
“Like, I think it is really interesting in that a lot of the, like for obvious reasons, a lot of the frontier labs, they're just not publishing anymore.”
Chen: Researchers Degrade AI Factuality Just to Boost LMSYS Rankings
“A lot of researchers, they'll tell us that their VPs make them focus on increasing their rank on LMSYS. And so I've had researchers explicitly tell me that they're okay with making their models worse. Add factuality works at following instructions as long as i…”