Bill Gurley and Sunny Madra talk open-source vs. proprietary AI | E1825

Episode Summary

Executive Summary: This podcast episode features Bill Gurley discussing the critical importance of open-source AI models versus closed-source approaches. The conversation highlights how leading AI companies like OpenAI and Anthropic are advocating for regulation to stifle competition, while open-source models like Llama 2 and Mistral are rapidly improving and gaining market share. The hosts argue that open-source AI fosters innovation, transparency, and societal benefit, warning against regulatory capture by incumbents.

Main Topics: Open Source vs. Closed Source AI (Priority: 5/5): Debate on the merits of open-source AI models versus proprietary ones, with emphasis on innovation, competition, and societal impact. Regulatory Capture in AI (Priority: 5/5): Discussion on how leading AI companies are pushing for regulation to protect their market position, echoing historical patterns of regulatory capture. Performance of Open Source Models (Priority: 4/5): Analysis of recent open-source models like Mistral and Lava, showing they are closing the gap with proprietary models in speed and capability. Cost and Energy Efficiency (Priority: 3/5): Examination of the high costs and energy consumption of large AI models, and how open-source alternatives can reduce these barriers. Historical Parallels and Lessons (Priority: 3/5): Comparison to past tech trends like Linux, Kubernetes, and the SBF case to illustrate risks of centralized control and benefits of open ecosystems.

Key Arguments: Open-source AI unleashes innovation and prosperity, while closed-source models stifle competition through regulatory capture. Incumbents like OpenAI and Anthropic are using safety concerns as a pretext to block open-source competition, similar to historical examples. Open-source models are rapidly improving, with Mistral's 7B parameter model outperforming larger models, demonstrating efficiency gains. Regulating open-source AI would put regions that adopt such regulations at a competitive disadvantage globally. The high cost and energy consumption of large models make open-source alternatives more accessible and sustainable.

Data Points: Mistral 7B parameter model performance: Outperforms Llama 2 13B - Demonstrates that smaller open-source models can compete with larger ones. Cost per ChatGPT query: 4 cents - According to Bernstein analysis, highlighting the expense of proprietary models. GPU cluster size for OpenAI: 10,000 GPUs - Indicates the massive infrastructure required for leading closed models. Energy consumption of H100 GPU: 700 watts - Illustrates the high energy costs of running large AI models. Time for Lava model to match GPT-4V: 1 week - Shows the rapid pace of open-source innovation.

Pivotal Quotes: "The people that are leading the charge, calling for the regulation and calling, and some of them raising this question of whether open source should be allowed, are the incumbents." — Bill Gurley: Highlighting the conflict of interest in regulatory calls. "You don't want any one party controlling a platform technology. You want it to be open source. You want closed source solutions, private company solutions, open source solutions. You want a range of opportunities." — Jason Calacanis: Summarizing the need for diversity in AI development. "The last time we had some founder say, trust me, I'll get us some regulation. That was SBF. And he was going to be the one who got us federal regulation for crypto. And he's on trial at this very moment for a bunch of mishugana." — Jason Calacanis: Drawing a cautionary parallel to the SBF case.

Implications: Listeners should advocate for open-source AI to prevent monopolistic control and ensure broad innovation. The rapid progress of open models suggests they will become increasingly viable, potentially disrupting current market leaders. Regulatory efforts by incumbents should be scrutinized for protectionist motives.