Oct 25, 2024 · 57m · sourcery
Former Chief Scientist at Salesforce, Richard Socher | You.com, LLMs, AI Agents, Complex Work · Sourcery with Molly O'Shea
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 Sourcery, host Molly O'Shea interviews Richard Socher, founder and CEO of You.com and managing director at AIX Ventures, following You.com's $50 million Series B funding round. Socher discusses You.com's technical architecture, enterprise growth, multiplayer team collaboration features, and his perspective on the AI venture landscape.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Molly holds 14.2% of the talking time here. How this is scored →
Molly as informed peer 2.6 Guest teaching 3.4 Guest disagreement 0.6 Molly pushing back 0.1
speaking balance: gold is Molly, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 39:40 Socher dismantles the 1-person AI unicorn narrative
Socher firmly rejects the speculative hype around single-person unicorns and pump-and-dump AI schemes, using mathematical compounding of error rates across agent steps to ground reality.
Hardest push from Molly ▶ 51:04 O'Shea challenges the sustainability of inflated AI valuationsO'Shea presses on whether current sky-high AI valuations are sustainable and challenges the industry logic behind pricing early-stage startups so aggressively.
Biggest teaching moment ▶ 23:15 Socher outlines the fundamental GenAI verification ruleSocher educates the host and audience with a structured framework: GenAI creates durable value when generation takes long for humans but verification is instant.
Molly holds their own ▶ 52:31 O'Shea drills into AIX Ventures' specific investment thesisO'Shea demonstrates her venture capital fluency by probing whether AIX's prominent portfolio is built on a proactive technical thesis or standard opportunistic network investing.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Molly as informed peer | Guest teaching | Guest disagreement | Molly pushing back | Why |
|---|---|---|---|---|---|---|
| Sourcery Show Title Sequence | 4 | 1 | 0 | 0 | Molly O'Shea introduces Richard Socher with detailed background on his career, You.com's Series B round, and investors. The exchange is purely welcoming and promotional. | |
| Evolution of You.com into a Knowledge Work Productivity Engine | 3 | 4 | 1 | 0 | O'Shea shares her personal experience using You.com and asks about its pivot. Socher explains the economic and consumer limits of basic web search and why knowledge workers require a productivity engine with verifiable citations. | |
| User Growth Metrics, API vs. Subscription Models, and Custom Agents | 2 | 4 | 0 | 0 | O'Shea asks how You.com offers all major LLMs at low cost. Socher details load balancing, orchestration, custom agents, and RAG architectures at scale. | |
| Sponsor Segment: Archer Aviation | 3 | 4 | 1 | 1 | After an ad read, O'Shea asks whether the battle in AI is primarily about accuracy versus flashy marketing features. Socher explains why accuracy drives enterprise sales after customers fail with hackathon prototypes. | |
| Technical Architecture: Search vs. LLM Stacks and Verification | 3 | 5 | 1 | 0 | O'Shea asks how You.com technically ensures factual accuracy. Socher breaks down the dual-stack model (search stack vs. LLM stack) and the core GenAI heuristic regarding artifact generation vs. verification time. | |
| Multiplayer Capabilities and Enterprise Team Collaboration | 2 | 2 | 0 | 0 | O'Shea asks about the new multiplayer feature. Socher explains default enterprise privacy and shared context diligence workspaces. | |
| The Vision for Autonomous Web-Surfing AI Agents | 2 | 4 | 0 | 0 | O'Shea asks about future agent capabilities. Socher frames agents as neural sequence models operating over action spaces, predicting autonomous agents will soon outnumber human web surfers. | |
| Enterprise Customer Profiles and Specialized Vertical Use Cases | 2 | 4 | 0 | 0 | O'Shea asks about ideal customer profiles. Socher gives technical examples from biotech analysis and insurance claim processing to illustrate high-value knowledge work. | |
| The Decreasing Marginal Cost of Intelligence and Economic Paradoxes | 3 | 4 | 1 | 0 | O'Shea brings up Socher's online post about falling marginal costs of intelligence. Socher invokes Jevons paradox, explaining that cheaper intelligence increases consumption rather than reducing demand. | |
| You.com's Trajectory, Revenue Growth, and Countering AI Hype | 2 | 4 | 2 | 0 | O'Shea asks if You.com could shrink to a 5-person company. Socher pushes back against the single-employee unicorn hype, pointing out cumulative error rates across multi-step agent workflows. | |
| Richard Socher's Founder Journey: MetaMind, Salesforce, and Stanford | 2 | 4 | 1 | 0 | O'Shea inquires about Socher's career thread. Socher recounts pushing neural networks for NLP in 2010 against academic skepticism, creating MetaMind, and formalizing prompt engineering in 2018. | |
| Evaluating AI Agent Startups: Investor and Operator Perspective | 3 | 3 | 0 | 0 | O'Shea asks how Socher evaluates agent startups as both operator and investor. Socher distinguishes between short-term leverage for junior programmers and long-term supervision by top engineers. | |
| AI Funding Landscape, Valuation Realities, and AIX Ventures Strategy | 4 | 3 | 1 | 1 | O'Shea questions the sustainability of sky-high valuations in AI seed rounds. Socher agrees, criticizing 'unicorn seed rounds' that pair early-stage risk with late-stage return profiles. | |
| Future Outlook: AI in Biology, Scaling You.com, and Hiring | 2 | 2 | 0 | 0 | O'Shea closes by asking what excites Socher over the next year. Socher highlights AI engineering in biology and gives hiring calls for You.com. |
Statements from this episode (20)
Socher: Georgian Led You.com's Series B Alongside NVIDIA and Salesforce
“The round was led by Georgian and participation from NVIDIA, Salesforce Ventures, several of the existing investors, Softlink Ventures Asia came into”
Socher: You.com was first to bring LLMs into search
“We're the first to bring LLMs into the search engine space and have some exciting patents some already given some pending on that”
Socher: AI search competitors fake citations and prioritize marketing over accuracy
“And when we do a lot of comparisons to the competition, usually the competition spends much more money on marketing and design and less on accuracy of the technology and accuracy of the answers and even the citations. A lot of folks just Fake their citations a…”
Socher: Several hedge funds and VC firms are enterprise You.com customers
“We actually have several hedge funds and VC firms that are actual paying customers and get their whole company on it because they understand how powerful it is.”
Socher: You.com receives more queries through its API than its consumer website
“We have millions of users now, which is really exciting to make many millions of calls a day. What's actually even more exciting is that we get an aggregate of more queries through our APIs than we get on even our own and User and user facing u.com property.”
Socher: You.com invented RAG at web scale before the term existed
“We realized just like with rag where we had invented rag already, we're doing it at a web scale, but it wasn't yet a term.”
Socher: Only 5% to 10% of enterprise AI seats get utilized
“This comes from a lot of organizations that bought
500 thousand plus seats for their product.
And then after a few months, they look into how many users are actually using it.
And it's like five, 10% in the company actually really use it.
And so, you know, the…”
Socher: Gemini's 2M token context window equals one 5MB file
“Google Gemini said, oh, we have a two million token context window, and that sounded amazing until you realize, like, that's one five megabyte file.”
Socher: GenAI thrives when creation is slow but verification is fast
“The longer it takes you to generate an artifact with Gen AI but the faster it is to verify if that artifact is actually useful and good, the better Gen AI can change that, that industry or that workflow.”
Socher: AI agents will outnumber human web surfers within years
“I think in the next few years, we're going to see more AI agents surfing the web for people than we'll have people surfing the web.”
Socher: Cheaper AI will dramatically increase total consumption via Jevons paradox
“As the cost of intelligence is going down. We're just going to use more intelligence. Everywhere, right? Like anything from your light switch to like your flight scheduling to your marketing automation, to your tier two support, everything will get more access…”
Socher: Most workers will become managers of AI agents
“That's what's going to happen when intelligence gets cheaper and cheaper. We'll just be all much more efficient. And most of us have to become managers of AI agents doing work for us.”
Socher: You.com revenue is doubling every quarter this year
“Revenues doubling every quarter this year and it's becoming really meaningful now.”
Socher: Single-employee unicorns will not emerge in next few years
“I don't think we'll have in the next few years, like the sort of unicorn with a single, single employee type of situation. I think that will probably take a little bit longer and we'll be non-trivial to get right.”
Socher: 95% accurate agents fail half the time over 15 steps
“If your agent is like 95% accurate on each step and you multiply .95, like times 15, then like half the time that process will fail, right? If it's not like implemented in a way that's very robust to errors in each step.”
Socher: No one was teaching neural nets for NLP globally in 2014
“No one was teaching neural nets for NLP in the world in 2014 when I graduated, and I thought, man, I'm still not, I can't convince some of these older professors, they're so, like sort of prefer other methods and don't believe in neural nets, so I got to train…”
Socher: Salesforce Research invented prompt engineering in 2018
“On the research side this was in 2018, we finally culminated all my ideas when we invented prompt engineering, and we didn't quite call it prompt engineering, but the basic idea was there, namely that you can prompt a single model with different questions, and…”
Socher: OpenAI cited Socher's 2018 paper, crediting it with influencing early GPTs
“That paper was actually cited by OpenAI, and they're still doing research papers. Like, we were told by some of the folks that worked on the first version of GPTs, like one and two, that that actually had influenced them to work more in NLP.”
Socher: AI Coding Agents Benefit Average Programmers More Than Top Coders
“At first, like a coding agent, for instance, right now is actually lifting up the at or below average programmers more and more. Like, if you're a solid programmer, but maybe not the best, like, you can get much more productive with an AI agent that helps you …”