Naveen Rao
Related people
Episodes
Notes
The Twenty Minute VC: Venture Capital | Startup Funding | The Pitch · 20VC: Why Robotics Will be 100x Bigger than Coding Agents |…
-
Why Is Naveen Rao Excited About AI?
Naveen Rao is personally pumped that AI is a tool and capability for asking much bigger questions about the universe. Many things in science stay unexplained because the problems were unapproachable: not enough humans could think about a problem and actually make progress. The surface area of problems people can go after and start to solve is increasing, and he finds that extremely exciting.
The idea that AI is going to kill all humans is the stupidest worldview he has ever heard, and he is completely tired of it. AI is what lets humans start to crack bigger problems. Exploring more of the universe depends on understanding deeper parts of physics. Wormholes and similar ideas have been theor
-
How Did Naveen Rao Feel Meeting Steve Jurvetson?
Naveen Rao walked into an early meeting with Steve Jurvetson nervous. He had heard Jurvetson was super smart and expected to be taken apart. The two started geeking out on numeric representations. Jurvetson marked up a couple of papers, asked questions, and kept a back-and-forth that felt like a conversation with a technical colleague. For Rao, it was one of the best investor meetings he had.
No one had pitched a hardware startup in 10 years in 2014. The work was so new that no one had pitched it yet. Jurvetson was almost champing at the bit to invest because he had been thinking about the same kind of company himself. The meeting ended in an immediate yes.
-
What Did Naveen Rao Change His Mind About?
Naveen Rao changed his mind in the last 12 months on the useful metric for progress in investing. Market validation is an old way of looking at it. For deep technical problems the market is already there, so the metric that matters is progress on solving the problem.
The old metric comes largely from SaaS. Market validation makes sense when there is no deep technology route and the company is building a product that serves a purpose. When a goal is almost true by inspection, as with curing cancer, there is no need to validate the market. The need is to validate progress on the technical problem.
Investors now fall into two camps. One camp invests heavily and treats valuation as secondary.
-
How Much Bigger Will Robotics Be Than Coding Agents?
Naveen Rao thinks robotics is going to be 100 times larger than agentic coding. The world is still scratching the surface of how to see the output of intelligence, and the physical world is a place intelligence has not entered at all.
Industrial robotics to date are not really intelligent. They are rote. A known way to do a task gets built in, and the machine repeats it, which is not novel problem solving. Physical problem solving arrives in the not distant future. Imagine billions of robots roaming around and solving problems dynamically. Something breaks, a robot cleans it up or replaces it. The cost of cleaning everything up, building new buildings, and solving physical problems will not
-
Why Does AI Leave Developer Budgets About the Same?
Naveen Rao says the share of developer salary that shifts to coding tools may stay roughly stable, because training people to judge good code is the hard part.
Benioff described spending $300 million of a $6 billion developer salary budget on core code and coding tools, the 5% case. The open question was whether the share stays at 5%, rises to 30% or 50%, or falls to 1%. Code quality matters more than lines of code. Very experienced developers, the 10X developers, build a sense of what looks good and what is aesthetically right, and they can make things work fast. Coders who know what is good verify model output and become more efficient. Pay for humans plus AI does not change drastically.
-
Are Pure Software Investors in Trouble?
Naveen Rao thinks hardware is the place to be over the next 10 years and is confident a pure software investor is in trouble. He does not think SaaS is going to be where margin is gained anymore, and he does not see software being a huge market in 20 years.
Hardware investing is quite hard and takes a very different mindset. The generation of investors now is not equipped to be hardware investors. A sugar high ran from the 2010 and 2011 timeframe until now. The pattern was to spin something up on an AWS cloud service, test the market quickly, try things out, and then go after a hundred million or $200 million market. SaaS was massive fragmentation of the market because the delivery mechanis
-
Is Reinforcement Learning Less Efficient Per Dollar?
Naveen Rao says reinforcement learning is more inefficient. Per unit of compute or per dollar, the model improves less, even though the training still keeps moving forward. He does not know that the shift toward it is more efficient. The aim is making models better at the tasks that matter.
For the next several years, he thinks the work is probably very important. Specialization will start in the middle layers and maybe even in the pre-training layer. He does not think the model companies are thinking very long term about performance, such as a massive breakthrough in how an AI system works that would give an advantage. In his view the big labs are not doing it as much.
One month late on a
-
Why Are Big AI Model Companies a Data Problem?
Naveen Rao says the big AI model companies are largely a data problem. Most of their capabilities have not come from advances in the model itself.
The market specializes, and working on important problems creates a big market that is defensible. A power law applies to skills: humans can learn many skills, yet some skills are more valuable than others, and the same pattern holds for data and curricula. Data work started as a commodity, a volume gathering of observations. Higher-quality data sets, including higher quality for coding and for chain of thought, now carry a lot of value. A model is almost just a manifestation of what it learned.
The data layer stays important going forward. Buil
-
Should I Trust AI Agent Decisions Yet?
Naveen Rao says people should not trust AI agent decisions as much as human decisions in a lot of cases just yet. The element that decides which solution is good is not well understood for every task.
Becoming a neurosurgeon takes a seven-year training after med school. Learning how to throw a stitch is not what makes the job hard, and he has done that in grad school. The physical, mechanical part of being a surgeon is not the problem. Judgment about which direction to go is the problem. It takes seeing enough cases, seeing the output, and seeing how people think through problems. Intelligence, as it is defined right now, starts with a stochastic model. Prompting it several times yields dif
-
Have Physical Efficiency Gains in Chips Stopped?
Naveen Rao says physical efficiency gains in chips have largely stopped. The last few generations of GPUs and other chips show little new efficiency at the physical layer. Moore's law and Dennard scaling made devices smaller, denser, and more efficient, and the innovations have largely stopped.
He splits how a computer is built into three levels. The physical level is the energy required to move a bit between two states, the way a phone's weight changes the energy needed to flip between zero and one. Most efficiency gains over the last several years, and over several generations, have come from the next layer, computing architecture, by moving to smaller data types: 32-bit, then 16-bit, the