By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16zWhat DeepSeek Means For The Future Of AI | Tech Veterans Weigh In
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
Tech Veterans Say DeepSeek Echoes the Internet, Not Sputnik
- Both guests see DeepSeek as a strong team's real, cumulative achievement, not a psyop or something from nowhere, while the roughly trillion-dollar market reaction was an overreaction.
- Casado highlights the permissive MIT-style license and the released reasoning traces as the most important features, since they enable distillation into small models and wide proliferation.
- Sinofsky frames the moment as a shift from scaling up to scaling out, comparing it to mainframes giving way to microcomputers, with value moving to specialized models on endpoints and to the app layer.
- Casado argues the biggest lesson is that US AI policy built on export controls and restrictions on open source was wrongheaded, and Sinofsky adds that the internet, not Sputnik, is the right analogy.
- Both are bullish on frontier labs and Nvidia, expect a much larger market, and advise labs to build apps themselves while startups win through workflow, stateful apps, and enterprise features.
IDEAS WORTH REMEMBERING
5 ideasDeepSeek was not a fluke or psyop
Casado and Sinofsky say it came from a top research team with about a year and a half of build-up, using techniques already in public literature, and that its roughly $6 million figure for the chain-of-thought work is not out of line with what Anthropic and OpenAI have said they spent.
Licensing and reasoning traces drive proliferation
Casado says the MIT-style license is rare for a state-of-the-art model, and releasing the chain of thought lets others distill small, high-quality models cheaply that can run on smaller devices.
The shift is from scaling up to scaling out
Sinofsky says big players default to bigger and bigger systems, as in mainframes versus microcomputers, but the real win comes from distributing computation across billions of endpoints with specialized models.
The AI buildout differs from the fiber glut
Casado notes that the dot-com fiber buildout was driven by a debt-laden company that collapsed, whereas AI is funded mainly by big cloud companies with hundreds of billions on their balance sheets, so a similar crash is unlikely.
Defensibility is moving to workflow and enterprise needs
Casado says models get copied quickly through distillation, so durable companies build stateful, retentive apps around them, and Sinofsky adds that enterprise features like single sign-on and controls become sticky.
WORDS WORTH SAVING
5 quotesIt's like this is another step to basically AGI in your pocket, right? Like, these can run on small models. It shows that we're going forward.
— Martin Casado
the biggest aha of DeepSeek is how blind our policies, our policies have been around AI, right?
— Martin Casado
the apps that ended up mattering on the internet literally didn't exist before the internet, and I think that's what people are losing sight of
— Steven Sinofsky
It's a good reminder that the, there are always pockets of people innovating.
— Steven Sinofsky
the lesson is not Sputnik, the lesson is the internet.
— Steven Sinofsky
High quality AI-generated summary created from speaker-labeled transcript.
