20VC: Foundation Models: Who Wins & Who Loses | How Economies and Labour Markets Need to Change in a World of AI | China vs the US in an AI Race: What You Need to Know | Rich Socher, Founder @ You.com

Unknown Source April 18, 2025 64 min

artificial-intelligence generative-ai startup ai-infrastructure investment google openai microsoft

🎧 Listen to Original

84 Companies

128 Key Quotes

5 Topics

2 Insights

🎯 Summary

20VC Podcast Summary: Foundation Models, Economics, and the AI Race with Rich Socher (You.com)

This 63-minute episode of 20VC features Harry Stebbings in conversation with Rich Socher, Founder and CEO of You.com, a pioneer in NLP who previously brought neural networks to the field and invented key concepts like word vectors and prompt engineering. The discussion centers on the current, confusing state of Foundational Models (LLMs), the resulting economic shifts, and the competitive landscape between the US and China.

1. Focus Area

The primary focus is the Foundational Model Layer (LLMs), covering their rapid commoditization, the shift from horizontal consumer applications to specialized enterprise solutions, the viability of advertising models in AI, and the competitive dynamics (US vs. China) in AI development. Secondary themes include the future of work, the nature of intelligence, and the challenges of building actionable AI agents.

2. Key Technical Insights

• Evolution of NLP: Socher traces his contributions from early word vectors to contextual vectors (leading to ELMo/BERT) and the initially rejected concept of Prompt Engineering, highlighting how foundational research often faces initial rejection before becoming obvious.
• Intelligence Upper Bounds: The discussion introduces a framework for categorizing intelligence into dimensions (e.g., object detection vs. knowledge), suggesting that while some areas are nearing their upper bound (solved), others, like comprehensive knowledge, remain astronomically far from AGI.
• Open Source Shock: The rapid emergence of high-quality, open-source models (like DeepSeek from China) demonstrated that achieving state-of-the-art performance might not require the multi-billion dollar budgets previously assumed, challenging the VC narrative around necessary capital expenditure.

3. Business/Investment Angle

• Commoditization of Infrastructure: Pure LLM infrastructure providers are likely to become commoditized, resembling telcos: high CapEx, high value creation for the world, but low value capture for the provider, as moats are weak (especially with open source).
• Value Capture in Enterprise: The real value capture is shifting away from horizontal consumer apps (where ChatGPT dominates) toward enterprise solutions that solve specific, complex internal workflows, as consumers often don’t need or won’t pay for complex AI interactions.
• Advertising Failure in AI: Socher asserts that ads in AI search/query interfaces perform 10x to 100x worse than traditional search ads, making the traditional Google revenue model difficult to replicate directly in generative AI interfaces.

4. Notable Companies/People

• Rich Socher (You.com): The expert guest, detailing his background and You.com’s pivot to enterprise solutions.
• OpenAI/ChatGPT: Acknowledged as the dominant consumer application, making it difficult for other horizontal LLM apps to gain traction.
• Anthropic: Mentioned as maintaining a consumer product (Claude) likely to prove model quality and engineering prowess.
• Google/Gemini: Discussed as facing the classic Innovator’s Dilemma—their core business model (ads on blue links) conflicts with providing the best, direct answers.
• DeepSeek: Highlighted as the open-source model that shocked the industry by matching or exceeding closed-source performance quickly.

5. Future Implications

• Agent Valley of Disillusionment: There is a current “valley of disillusionment” regarding action agents (AI that takes irreversible actions like booking flights) because they lack the deep, subtle, personalized knowledge about the user required for complex tasks.
• Unbundling Wave: The industry is currently in a large wave of unbundling, where users move away from monolithic search (Google) to specialized apps (Windy for weather, TikTok for discovery, Amazon for small purchases). LLMs will capture the complex query segment of this unbundling.
• Enterprise Adoption Hurdles: Large enterprises struggle with adoption (paying for licenses that go unused) because employees, used to being individual contributors, are not yet skilled at distilling knowledge into succinct, unambiguous prompts for agents.

6. Target Audience

This episode is highly valuable for Venture Capitalists, AI/ML Engineers, Enterprise Software Strategists, and Technology Executives who need a nuanced, expert perspective on the current market dynamics, investment viability, and technical trajectory of foundational models beyond the hype cycle.

🏢 Companies Mentioned

Linux ✅ technology_analogy

Instagram ✅ technology_analogy

Flickr ✅ technology_analogy

LLM4 model ✅ ai_model

Apple ✅ big_tech

DoorDash ✅ technology_company

Uber Eats ✅ technology_company

Windy ✅ technology_company

Yelp ✅ technology_company

TikTok ✅ technology_company

Amazon ✅ big_tech

Waze ✅ technology_company

Mode Mobile ✅ technology_company

Brex ✅ ai_application

Notion ✅ ai_application

💬 Key Insights

"What I'm really excited about is getting quantum computers to a scale where we can simulate a cell... to from real first principles like really deeply model physics and chemistry that is complex and eventually biology. Like that will be such a massive unlock because in AI anything you can simulate AI can solve every problem in that domain."

Impact Score: 10

"LLMs are garbage in, garbage out to a large degree. So if you have a search model or an index, and you ask, like, what's new with Trump? And that search index brings back pages from like six years ago, the LLM will tell you wrong and outdated things about that query. So the search is kind of the often forgotten infrastructure layer for LLMs."

Impact Score: 10

"I think the biggest problem and biggest worry I have is that the entry-level jobs right now are more and more automatable. And so you hopefully have companies that have a long enough time horizon, such that they're willing to train people even though any AI could do the job..."

Impact Score: 10

"Knowing how to program is incredibly important. Why do you say that with the commoditization of a lot of low-level programming? For the same reason why we're speaking English... But programming isn't just about programming itself... it's also about a different way of thinking. And it's a different way of understanding the world that you're in."

Impact Score: 10

"But assuming back in, I think the problem is that while most people complain about their jobs, it does give them meaning. It does give them meaning to be a valuable part of society and to have earned something that they can then give to their family, to their kids and so on. And so I think UBI will take that meaning away and we're already in a meaning crisis with the technology and society that we have set up for ourselves in many places and I think that was just exacerbated."

Impact Score: 10

"AI is the perfect tool to tackle that kind of complexity because we can now have similar things. We understand how one neuron works but when you have enough of them, you scale it up enough. All of a sudden you can have a conversation and people think it's a human being on the other side."

Impact Score: 10

📊 Topics

#artificialintelligence 158 #generativeai 32 #startup 16 #aiinfrastructure 5 #investment 4

🧠 Key Takeaways

💡 think about it

💡 like spread consciousness into the universe