Oct 10, 2024 · 1h 28m · mad
AGI, The Future of AI Agents And The Next Wave of Opportunities in AI | Richard Socher, CEO, You.com
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 The MAD Podcast, host Matt Turck interviews Richard Socher, CEO of You.com and Managing Partner at AIX Ventures, exploring the macro impact of artificial intelligence, scaling laws, and AI agent architectures. Socher shares insights on enterprise AI deployment, You.com's productivity engine, frontier startup investing, and historical paradigms shaping the future of technology.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 14.7% of the talking time here. How this is scored →
Matt as informed peer 4.5 Guest teaching 4.7 Guest disagreement 1.9 Matt pushing back 1.8
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 48:15 Dismissing staged AI agent video demos
Socher forcefully dismisses polished AI agent demos showing seamless travel booking, calling out marketing exaggeration by stating 'no way in hell that was real'.
Hardest push from Matt ▶ 50:40 Challenging guest on Salesforce's enterprise agent advantageTurck explicitly challenges Socher's view on incumbent agent capabilities, pointing out that Salesforce possesses massive amounts of native workflow action data through Agentforce.
Biggest teaching moment ▶ 32:30 Explaining multi-dimensional intelligence and metacognition boundsSocher provides a detailed educational breakdown of the dimensions of intelligence, explaining why current LLMs lack self-reflection and objective-setting metacognition.
Matt holds his own ▶ 24:00 Citing neuro-symbolic and post-transformer architecturesTurck demonstrates strong technical domain depth by introducing neuro-symbolic AI and asking whether post-transformer state-space models like Mamba will succeed current architectures.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Trailer and Episode Highlights | 0 | 0 | 0 | 0 | Host monologue intro and teaser clips setting up the episode; standard podcast opening with no active discussion dynamics. | |
| Productivity Leaps and Automating Intellectual Work | 2 | 4 | 1 | 1 | Socher outlines the historical parallel between the Industrial Revolution and AI productivity leaps, while Turck jokingly asks if his email outreach could be automated. | |
| Skill Disruption and Abstract Work Hierarchies | 3 | 5 | 1 | 1 | Socher explains how lower-performing workers benefit most initially from AI tools and how work moves to higher abstraction levels, with Turck asking what defines top performers in that regime. | |
| Democratizing Elite Services and Modern Polymaths | 4 | 4 | 1 | 1 | Turck introduces the concept of the modern Renaissance polymath, which Socher expands upon by detailing how elite services like personal tutors will be democratized. | |
| Deep Learning Persistence and ImageNet Contributions | 4 | 4 | 2 | 1 | Turck brings up Socher's early days in deep learning research and the Jevons paradox, prompting Socher to recount pushing through early academic skepticism regarding neural vectors. | |
| Scaling Laws and Multimodal Neural Sequence Models | 5 | 6 | 2 | 2 | Turck frames the scaling law bottleneck debate clearly, prompting Socher to break down text data limits, code-driven logical reasoning, and multimodal sequence modeling. | |
| Model Architectures, Neuro-Symbolic AI, and Value Capture | 6 | 5 | 2 | 2 | Turck demonstrates solid domain depth by inquiring about neuro-symbolic AI and post-transformer architectures, while Socher uses a telecom infrastructure analogy to discuss value capture. | |
| Defining AGI, ASI, and Dimensions of Intelligence | 4 | 7 | 2 | 2 | Socher delivers an extensive multi-dimensional definition of intelligence, explaining to Turck why superhuman capabilities in narrow domains differ from self-aware metacognition. | |
| Hype Cycles, Production Realities, and Mild Winters | 5 | 5 | 3 | 2 | Socher rejects extreme marketing claims by comparing current LLM expectations to early self-driving car hype, while Turck highlights that baseline capabilities have passed the point of no return. | |
| Advanced RAG and Verifiable Citations in Enterprise AI | 5 | 6 | 3 | 2 | Turck seeks technical clarification on RAG systems, leading Socher to expose competitor flaws like hallucinated or irrelevant citations. | |
| Autonomous AI Agents and Web Infrastructure Impact | 5 | 6 | 3 | 2 | Turck asks for clear definitions of AI agents, prompting Socher to explain action sequence models and detail the compound mathematical error rates that break multi-step agent workflows. | |
| Enterprise Data Constraints and the Productivity Engine Category | 6 | 5 | 3 | 4 | Turck pushes Socher on whether incumbents like Salesforce have a structural advantage with enterprise agent data, with Socher arguing that tenant isolation prevents global data aggregation. | |
| You.com's Expansion, B2B APIs, and Monetization Strategy | 5 | 5 | 2 | 3 | Turck points out that running consumer, enterprise, and API businesses simultaneously is usually discouraged by VCs, which Socher acknowledges while defending their market pull. | |
| Multi-Model Routing and Eliminating Vendor Lock-in | 5 | 4 | 1 | 1 | Turck asks about the internal mechanics of You.com's model orchestration layer, allowing Socher to explain how abstracting multi-model vendor lock-in creates customer value. | |
| Prioritizing Accuracy over Cost and the LLM OS Framework | 6 | 5 | 2 | 3 | Turck presses on unit economics and token costs, leading Socher to articulate why enterprise accuracy matters far more than optimizing raw token expense. | |
| Deep System Architecture and Multi-Turn Query Rewriting | 5 | 6 | 2 | 2 | Turck asks what differentiates You.com when using third-party LLMs, prompting Socher to explain intent classification and multi-turn query rewriting sub-modules. | |
| AIX Ventures and Venture Capital Fund Strategy | 4 | 3 | 1 | 1 | Turck asks about AIX Ventures' structure, allowing Socher to highlight early investments like Hugging Face at a five million valuation and their domain-expert model. | |
| Evaluating AI Startups, Climate Tech, and Biotech Innovations | 5 | 5 | 3 | 2 | Turck questions valuation sanity in early-stage AI, with Socher warning against taking seed-stage risk at late-stage prices before sharing non-traditional investments in climate tech and biotech. | |
| AI Researchers as Founders and Perspectives on OpenAI | 6 | 4 | 2 | 3 | Turck brings up the tensions surrounding AI researchers becoming founders and OpenAI's structural changes, with Socher offering a balanced perspective on founder conviction. |
Statements from this episode (33)
Turck: Socher published 'The Age of AI' weeks before Altman's manifesto
“You wrote recently a blog post that you called the age of AI. And I have to say that was like, At least four weeks before Sam Altman published his own manifesto on the intelligence age.”
Socher: Testing students' writing requires removing internet and using paper
“If you want to test how students write essays, you probably have to take away the internet and just have them do some handwritten essays sometimes if that's still what you want to test for.”
Socher: Lower-performing workers benefit most from initial AI automation
“In the beginning of this industrial revolution and workflow sort of process automation that we're seeing, the bottom half of performers actually benefit the most.”
Socher: Top programmers get limited AI benefits but provide its training data
“Currently, the top programmers don't actually benefit that much from AI, but they're informing AI a lot, and they're giving better training data to it.”
Socher: AI will give every child a personal tutor within years
“Now, what do really wealthy people currently have where AI still hasn't done that yet, but we will obviously in the next few years. It's like a billionaire can have a personal tutor for their kids. AI will be a personal tutor for your kids.”
Socher: I was a co-author of the seminal ImageNet paper
“And yeah, I was one of the coauthors of ImageNet too.”
Socher: Cheaper AI will skyrocket intelligence usage across tasks and personal assistants
“As intelligence gets cheaper and cheaper, We won't just lose all the jobs that currently require intelligence. We will just use intelligence in more and more places. I think overall, the amount of, like, tasks that, you know, we believe require some decent amo…”
Socher: AI scaling will hit logarithmic leveling off in reasoning gains
“I do think we are going to hit a sort of logarithmic kind of leveling off off like how much more data will give us that much more reasoning, right?”
Socher: Top AI models have run out of high-quality human text data
“I think a lot of the top models, you can't add multiple orders of magnitude more text to them, because there isn't that much more text around.”
Socher: Llama 4 will handle many enterprise AI workflows open source
“Certainly, I think once Lama IV comes out, a lot of the workflows can be, like, done with an open source Lama IV model.”
Socher: Scaling laws will show major progress in non-text modalities
“I think we're going to see incredible progress in scaling laws in just these new modalities.”
Socher: Transformer results would take 10x compute and engineering with LSTMs
“Probably if it wasn't for transformers, it would have just been like 10 X more engineering and data needed to get to similar results, even with like past models like LSTMs and so on.”
Socher: Open-source models will make value capture harder for foundation model companies
“Once others can go out into the world and an open source, a massive, large language model it's gonna be harder and harder to capture that value.”
Socher: Nvidia will match AI hardware startups' 100x gains before startups reach scale
“By the time you get that out and it's actually scalable and it can really train it and all the software is ready for it. And I can now go on AWS and spawn up your new hardware, like, Nvidia will also have been a hundred X faster with their latest and greatest …”
Socher: No company has made significant progress in conscious, self-aware AI
“There is no company that has made any significant progress in conscious, self-aware, self-reflective AI. And hence, it's very hard to predict when it will ever happen because we're making no progress towards it.”
Socher: There will never be another severe AI winter
“There will never be an AI winter again, the way we've seen in the past. If at all there will be an AI winter, it'll be like a California type winter.”
Socher: You.com was the first search engine to provide citations for LLMs
“We actually had the first LMs in u.com, like, in twenty-twenty-one already, filed some interesting patents on that and Then we were the first to actually, in a search engine context, give them citations, right?”
Socher: Generative AI disrupts fields where verification is faster than creation
“Whenever it's very slow to create an artifact in your space, an image, a text document, and so on, very slow to create it, but very quick to verify its correctness, then Gen AI is going to, like, massively disrupt that space.”
Socher: Half of competitors' AI search citations are random, irrelevant links
“Some of our competitors, half of their citations are Independent, random links that have nothing to do with the sentence that they're behind.”
Socher: Web traffic from AI agents will surpass humans within years
“I think in the next few years, we're going to see more AI agents surfing the web. Then people surfing the web”
Socher: 95% step accuracy yields 50% failure rates in 15-step AI tasks
“Even if your agent is 95% accurate, but now it's doing, like, 14, 15 steps, each of which is 95% accurate, and you multiply, you know.951549, now, half the time, the overall sequence is wrong, right?”
Socher: Salesforce is uniquely positioned to automate repetitive enterprise workflows
“My hunch is over the next few years more and more of that workflow, if it feels very repetitive, Employees are going to be like, I want to work in a company where if I give an X, 10 examples of a workflow to my tools, I expect that tool to then automate that f…”
Socher: Startups cannot be 10x better than Google on simple consumer search
“If you think about it from first principles, there is nothing you can do to be 10 X better.”
Socher: You.com gets more queries through APIs than consumer product
“We definitely get a lot more queries coming in through APIs than we do over like the normal consumer you.com product at this point.”
Socher: Pursuing consumer and enterprise motions simultaneously is generally a bad idea
“It's generally a bad idea. I would discourage everyone from doing it. You need a lot more extra funding. You need a lot of very smart people who execute at extremely high levels. It makes everything harder.”
Socher: Anthropic's Claude models excel at legal reasoning tasks
“One very concrete example is Anthropics, quite good for legal types of reasoning and like just kind of connected to some of their morals and ethics and so on. And so that those questions are often better routed to a Claude-like model from Anthropics.”
Socher: You.com pioneered routing LLM queries to non-language UI widgets in 2023
“That idea, we were the first to launch that early, 20, 23.”
Socher: I angel invested in Hugging Face at a $5M valuation
“I invested in these two three founders creating this cute company called Hugging Face. I had a beautiful five million dollar valuation. And now they're worth, like, four and a half billion, right?”
Socher: AI research papers predict future commercial AI capabilities
“You can actually, if you want to predict like, oh, what's the next thing of open AI? Just read all the research papers over the last couple of years. And a lot of those will come and get into products at scale with all the right engineering and so on. And then…”
Socher: AIX Ventures hit 3x TVPI on its first venture fund
“I put all my angel portfolio into Fund One, and we raised Fund Two now, a two hundred two million dollar fund, and Yeah. Already have like three X TVPI or fund one.”
Socher claims Coca-Cola is the world's largest milk producer
“Coca-Cola, which is actually the largest milk producer in the world, a lot of people don't know that, but they do a lot of things”