The Twenty Minute VC (20VC) Episode Summary: Deepseek, AI Arms Race, and the Future of Inference with Jonathan Ross

Podcast Information:

• Title: The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
• Host: Harry Stebbings
• Guest: Jonathan Ross, Co-founder and CEO of Groq
• Episode: Deepseek Special: Is Deepseek a Weapon of the CCP | How Should OpenAI and the US Government Respond | Why $500BN for Stargate is Not Enough | The Future of Inference, NVIDIA and Foundation Models with Jonathan Ross @ Groq
• Release Date: January 29, 2025

1. Introduction to Deepseek and Jonathan Ross's Expertise [04:01 - 05:00]

Harry Stebbings welcomes Jonathan Ross, CEO of Groq, highlighting Ross's extensive background in AI, including his pivotal role in Google's TPU initiative. The discussion centers around the recent news about Deepseek, positioning it as a significant development in the AI landscape.

Notable Quote:

Jonathan Ross [05:08]: "Yes, it's Sputnik. It is Sputnik 2.0. Even more so."

2. Deepseek's Breakthrough and Competitive Edge [05:00 - 10:00]

Ross delves into why Deepseek is making headlines, comparing its advancement to Sputnik 2.0. He explains that unlike traditional models that rely heavily on scaling with more GPUs, Deepseek achieved superior performance with a tighter budget and innovative techniques.

Key Points:

• Training Efficiency: Deepseek trained its model on approximately $6 million worth of GPUs, comparable to OpenAI's Llama 70 model but achieved better results.
• Data Quality Over Quantity: Emphasizes the importance of high-quality data and reinforcement learning from OpenAI to enhance model performance without massive computational resources.
• Innovative Distillation: Deepseek employed automated methods to refine their model, bypassing the need for extensive human intervention.

Notable Quote:

Jonathan Ross [07:16]: "It's a little bit like speaking to someone who's smarter and getting tutored by someone who's smarter."

3. Implications for OpenAI and the US Government [10:00 - 16:00]

The conversation shifts to potential responses from OpenAI and the US government to Deepseek's advancements. Ross speculates whether OpenAI might open-source their models to maintain competitiveness and user trust.

Key Points:

• Open-Source Advantage: Ross suggests that OpenAI could benefit from open-sourcing their models to retain user base and goodwill.
• Export Control Loopholes: Discusses the challenges in enforcing export laws on GPU usage, noting that companies can easily rent GPUs from cloud providers, undermining restrictions.
• Data Security Concerns: Highlights the risks of user data being accessed by the Chinese Communist Party (CCP) through models like Deepseek, raising alarm over data privacy and security.

Notable Quote:

Jonathan Ross [12:59]: "Do you really think like the CCP doesn't have all your data and isn't going to look it up later?"

4. The AI Arms Race: US vs. China [16:00 - 28:00]

Ross elaborates on the broader geopolitical implications of AI advancements, particularly the competitive dynamics between the US and China.

Key Points:

• China's Dominance in AI: China's aggressive investment and resource allocation in AI place them at a significant advantage in the global AI arms race.
• European Challenges: Europe struggles with risk aversion and lacks the agility seen in the US and China, potentially hindering its competitiveness in AI.
• Mutually Assured Destruction Paradigm: Ross draws parallels to the Cold War, suggesting that without strategic measures, the AI race could lead to unchecked escalation.

Notable Quote:

Jonathan Ross [28:00]: "If you're running in a race with someone who's willing to take steroids, if you want to win, you're going to have to take steroids too."

5. Future of AI Compute: Training vs. Inference [28:00 - 38:16]

The discussion transitions to the technical aspects of AI compute, differentiating between training and inference, and their respective impacts on the industry.

Key Points:

• Inference Dominance: Ross predicts that inference will account for up to 95% of AI compute spending, as models are primarily used rather than continuously trained.
• Mixture of Experts (MOE): Explains Deepseek's use of MOE to enhance efficiency by activating only relevant model parameters, akin to selective neuron firing in the human brain.
• Nvidia's Role: Highlights how advancements in inference technology bolster Nvidia's market position, despite short-term market fluctuations.

Notable Quote:

Jonathan Ross [20:36]: "I can't tell you about the popularity contest, but in terms of the weighing machine part, this is a misunderstanding. It's actually more valuable, thanks to Deepseek not less valuable."

6. Marketing, Brand Power, and the Seven Powers Framework [38:16 - 44:38]

Ross introduces Hamilton Helmer's Seven Powers framework to analyze the competitive strategies of top tech companies in the AI space.

Key Points:

• Seven Powers: Discusses how marketing serves as a tool to decommoditize products, with brand power being a critical factor.
• OpenAI's Strategic Positioning: Suggests that OpenAI should leverage its brand power and consider open-sourcing to maintain its competitive edge.
• Meta's Advantage: Due to their network effects, Meta inherently benefits from open-source advancements, making them formidable competitors in the AI arena.

Notable Quote:

Jonathan Ross [18:25]: "Marketing is the art of decommoditizing your product. And the seven Powers are seven great ways to decommoditize your product."

7. The $500 Billion Stargate Initiative [44:38 - 45:09]

Harry inquires about OpenAI's ambitious $500 billion Stargate plan. Ross provides his perspective, emphasizing that infrastructure investment alone may not suffice without strong brand power.

Key Points:

• Infrastructure Challenges: Building out large-scale AI infrastructure requires substantial capital and time, making rapid scaling difficult.
• Brand Over Infrastructure: Ross argues that OpenAI should prioritize brand strengthening to complement infrastructure investments, ensuring sustained competitiveness.

Notable Quote:

Jonathan Ross [44:36]: "The real win here is brand. That's what I would be doubling down on."

8. The Road Ahead: Innovation, Regulation, and Market Dynamics [45:09 - 51:55]

In the concluding segments, Ross reflects on the future trajectory of AI, encompassing innovation potentials, regulatory landscapes, and market adaptability.

Key Points:

• Continuous Disruption: The AI industry is poised for relentless innovation, with companies needing to pivot and adapt to survive.
• Regulatory Imperatives: Suggests that governments must establish clear, automated responses to unfair practices like subsidization and intellectual theft to maintain a balanced competitive environment.
• Ethical and Security Concerns: Raises alarms about AI being used for cyberattacks and the difficulty in attributing such malicious activities, potentially escalating geopolitical tensions.
• Opportunities in High-Quality Products: Despite model commoditization, there remains a significant market for well-crafted, high-quality AI-driven products that excel in user experience and reliability.

Notable Quotes:

Jonathan Ross [50:41]: "I think people will always prefer to use the highest quality, most polished product."

Jonathan Ross [47:23]: "The thing that I am most nervous about is that unlike nuclear war, you can use AI tools to attack each other."

Conclusion

This episode of The Twenty Minute VC provides an in-depth analysis of Deepseek's recent advancements and their broader implications on the AI industry and global geopolitics. Jonathan Ross offers expert insights into the competitive dynamics between major tech players, the critical importance of inference in AI compute, and the strategic maneuvers necessary for companies like OpenAI to maintain their dominance. The discussion also underscores the urgent need for balanced regulatory frameworks to prevent unchecked AI-driven competition and safeguard data security.

For listeners interested in the intersection of venture capital, technology, and global strategy, this episode offers valuable perspectives on navigating the rapidly evolving AI landscape.