DeepSeek’s Jaw-Dropping Performance and the Future of AI

A Conversation with Anjney "Anj" Midha

Andreessen Horowitz general partner and Mistral board member Anjney "Anj" Midha has been following the rapid progress of DeepSeek, an AI chatbot app that has been making waves in the tech industry. In a recent conversation with TechCrunch, Midha shared his insights on DeepSeek’s impressive performance and its potential impact on the future of AI.

DeepSeek’s Breakthrough

Midha first spotted DeepSeek’s impressive performance six months ago, when the company introduced Coder V2, a coding-specific AI model that rivaled OpenAI’s GPT4-Turbo. This breakthrough put DeepSeek on a path to release improved models every couple of months, including its new open-source reasoning model, R1.

R1: A Game-Changer for AI

R1 is a game-changer for AI, offering industry-standard performance at a fraction of the cost. Midha believes that R1 will enable companies to do more with the compute power they can obtain, rather than simply spending billions on GPU chips and data centers.

The AI Hierarchy

Midha notes that the AI hierarchy is not just about the amount of money spent on AI, but also about the efficiency of the models. He argues that Mistral’s open-source approach gives it an edge over closed-source rivals, which have to pay for all the labor and compute power.

The Competition

Despite the sell-off of Nvidia’s stock, Midha believes that the AI industry will continue to grow, with companies like Facebook’s Llama and OpenAI’s GPT4-Turbo continuing to invest heavily in AI research and development.

a16z’s Oxygen GPU Sharing Program

Midha is also the leader of a16z’s Oxygen program, which provides private GPU clusters to portfolio companies. The program is "overbooked" right now, with not enough GPUs to meet the demand of his startups.

The Future of AI

Midha believes that DeepSeek’s engineering breakthroughs will not change the fact that Stargate, OpenAI’s big $500 billion partnership with SoftBank and Oracle, is still on the horizon. However, he does think that DeepSeek’s recognition by nation-states as a foundational infrastructure, like electricity and the internet, will lead to a shift towards infrastructure independence.

The Concerns

Midha is concerned about the risks of Chinese models, with its censorship and claws in data. He believes that Western nations should use Western models, like Mistral, which follow Western laws and ethics and abide by NATO agreements.

The Reality

Not everyone shares Midha’s concerns, and companies can run Chinese open-source models locally in their own data centers. DeepSeek is also available as a secure cloud service from American companies like Microsoft Azure Foundry.

The Conclusion

In conclusion, DeepSeek’s jaw-dropping performance has significant implications for the future of AI. While the competition is fierce, Midha believes that Mistral’s open-source approach gives it an edge. He also notes that the AI hierarchy is not just about the amount of money spent, but also about the efficiency of the models. As the AI industry continues to grow, it’s clear that DeepSeek will play a significant role in shaping its future.