In this episode of Augment AI, hosts Ayesha Khanna and Bernard Leong provide an insightful analysis of DeepSeek, the Chinese AI model disrupting the global AI landscape. They trace DeepSeek's journey from its beginnings as a side project of hedge fund manager Liang Wenfeng to becoming a formidable competitor to OpenAI and Anthropic. Ayesha and Bernard break down the technical innovations behind DeepSeek's efficiency, discussing how its model distillation, mixture of experts approach, and multi-hit latent attention techniques dramatically reduce computational requirements. They debate whether DeepSeek's claimed $6 million training cost (versus billions spent by US companies) is accurate, with Bernard estimating the true cost between $35-70 million—still significantly cheaper than competitors. The hosts highlight DeepSeek's dramatic price advantage: $2.19 per million tokens versus OpenAI's $60. Ayesha shares perspectives from her global network in Europe, India, and Southeast Asia about how DeepSeek is democratizing AI access worldwide. Bernard explains what "open source" truly means in AI, while both hosts discuss the emerging copyright challenges in AI development and how companies worldwide are responding to DeepSeek's breakthrough that's reshaping industry expectations around cost, efficiency, and innovation speed.

Episode Highlights:
[00:00:57] Ayesha Khanna introduces her background in machine learning, Wall Street experience, and AI engineering firm in Singapore.
[00:03:22] Bernard Leong shares his background as a theoretical physicist and experience in machine learning, genomics, and corporate roles.
[00:06:50] DeepSeek's origin story as a side project of hedge fund manager Liang Wenfeng in China in late 2023.
[00:08:51] DeepSeek R1 model rivaled OpenAI's models and caused NVIDIA's stock to drop by $560 billion.
[00:10:53] Discussion of DeepSeek's rapid development speed challenging tech giants' timelines.
[00:13:10] Cost comparison: DeepSeek charges $2.19 per million tokens vs OpenAI's $60, with self-hosting being free.
[00:15:01] Details about the talent behind DeepSeek - young PhDs from Zhejiang University costing 1/5 of US counterparts.
[00:17:58] Explanation of Mixture of Experts model that activates only 37 billion parameters instead of 671 billion.
[00:19:25] Technical innovations including Group Relative Policy Optimization and multi-hit latent attention.
[00:21:50] Debate on whether DeepSeek's $6 million training cost claim is accurate (estimated $35-70M).
[00:24:28] Global impact of DeepSeek democratizing AI access for emerging markets.
[00:25:37] Discussion of what "open source" truly means in AI models and what DeepSeek actually shares.
[00:28:39] Conversation about copyright issues in AI model training.
[00:30:57] DeepSeek's open source week with five repositories improving inference speed and efficiency.
[00:34:12] Importance of inference speed in user experience and competitive impact in China.
[00:38:08] Hosts compare their usage of multiple AI assistants and getting spoiled by speed expectations.
[00:39:39] Connection between hedge fund efficiency mindset and AI model optimization.
[00:41:41] Concluding remarks and preview of future episode topics.

Main Site: https://www.theaugment.ai/

LinkedIn Page: https://www.linkedin.com/company/augmentaipodcast/

In this episode, tech analyst Benedict Evans joins hosts Ayesha Khanna and Bernard Leong to explore the rapid evolution of AI. They discuss the scale of AI investment, its real-world applications, and whether generative AI is truly a revolutionary shift or just another tech hype cycle. Benedict breaks down how businesses are adopting AI, the challenges of integrating it into workflows, and the broader impact on industries and jobs. With insights on scaling laws, open-source vs. closed models, and the future of automation, this conversation unpacks AI’s role in reshaping the world as we know it.

Episode Highlights
[00:00] Introduction of hosts and podcast
[01:45] Benedict's background in tech analysis
[03:00] Evolution of AI definition and perception
[05:30] Benedict's personal AI usage patterns
[07:30] Overview of "AI is Eating the World" presentation
[10:45] Patterns of technology deployment and adoption
[13:30] Understanding scaling laws in AI development
[16:15] Economics of foundation models
[20:00] Current state of enterprise AI adoption
[24:45] Evolution of AI interfaces: chatbots vs. embedded APIs
[27:30] Open source vs. closed source AI models
[31:00] Domain-specific models for specialized industries
[35:30] AI in context of broader technological landscape
[39:00] AI's impact on robotics development
[42:00] AI applications in military and defense technology
[48:00] Managing AI error rates as probabilistic systems
[55:30] Navigating polarized perspectives on AI
[59:00] AI and job displacement considerations
[01:05:00] Consumer AI product-market fit challenges
[01:09:30] Metrics for tracking AI adoption and impact
[01:12:00] Closing remarks and where to find Benedict online

Augment Main Site: https://www.theaugment.ai/

Augment LinkedIn Page: https://www.linkedin.com/company/augmentaipodcast