Latest Interviews
Showing 1–5 of 5 transcripts.
Clear all filters- a16z27 min
DeepSeek, Reasoning Models, and the Future of LLMs
Guido Appenzeller, Marco Mascorro
DeepSeek R1 is an open-weight reasoning model from China that achieves top-tier performance by combining Multi-Head Latent Attention, Group Relative Policy Optimization, and a 256-expert MoE architecture to generate complex thought chains. The development team overcame early behavioral failures through a low-cost, self-supervised pipeline utilizing 800,000 verifiable traces and rule-based verification to produce responses up to 10,000 tokens long for roughly $5.5 million in base training costs. This breakthrough has shifted industry focus toward test-time compute and local deployment, enabling state-of-the-art reasoning on consumer hardware while bypassing traditional bottlenecks associated with human-labeled data.
- a16z39 min
Two-Sided Networks in Healthcare - a Founder's Playbook
Vineeta Agarwala, Jay Rughani, Vinita, Arif Nthu, Andrew Adams, Iman Abouzeid, Thomas Closel, Doug Hirsch
This event explores the architecture of two-sided healthcare networks that connect the "five P's" of the industry to bridge fragmented data silos and create self-reinforcing value loops. Founders and investors detail specific go-to-market strategies, such as prioritizing demand or supply first, while emphasizing the critical need for early trust-building and the rejection of vanity metrics in favor of qualitative product validation. The discussion further outlines frameworks for scaling to a second side, maintaining network neutrality to prevent disintermediation, and evaluating defensibility against direct transactions as healthcare data infrastructure expands.
- a16z21 min
AI is Industrializing Discovery
The speaker characterizes the current biotech era as an industrial revolution where AI is transforming drug discovery from a bespoke artisanal model into a scalable, engineering-driven process. By overcoming myths regarding biological complexity and data scarcity through techniques like graph convolutions and one-shot learning, companies are achieving accuracy rates of over 90% in biomarker discovery and predicting clinical outcomes with modest incremental gains. This strategic shift promises to slash manufacturing costs for protein therapeutics by up to 90% and reduce trial expenses, creating a critical opportunity for professionals to merge domain expertise with AI skills to industrialize discovery at unprecedented scale.
- a16z27 min
The Architecture of Crypto Innovation
Addressing historic distrust in traditional financial institutions, Satoshi Nakamoto introduced Bitcoin as a decentralized system that solved the Byzantine consensus problem by requiring proof of participation cost rather than identity verification. The speaker analyzes the trade-offs between various incentive models like Proof of Work and Proof of Stake while detailing how technical, legal, economic, and societal factors create a complex spectrum of decentralization. Finally, the discussion highlights current scalability challenges, noting that increasing transaction throughput often necessitates sacrificing decentralization or relying on layered solutions like sharding and off-chain networks.
- a16z22 min
The Economics of Term Sheets
This presentation dissects economic term sheet mechanics using hypothetical venture firms Haiku and Indigo to illustrate how varying liquidation preferences and option pools directly impact founder ownership. By contrasting Haiku's participating 1x preference with Indigo's non-participating structure, the analysis demonstrates that while Indigo offers less immediate dilution, Haiku's smaller capital injection and option pool size yield a different risk-reward profile for the founding team. The session concludes by framing the choice between these competing $2 million versus $4 million offers as a strategic calculation of runway extension versus long-term upside, pending a future discussion on governance rights.