Jan 2, 2019 · 28m · a16z
a16z Podcast | A New Lab Rises
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
UC Berkeley professor and Databricks co-founder Ion Stoica joins the a16z podcast to discuss the transition from AMPLab to RISELab, the power of open-source software, and how university research bridges academic innovation with enterprise technology.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 21.3% of the talking time here. How this is scored →
The host as informed peer 4.6 Guest teaching 5.3 Guest disagreement 0.6 The host pushing back 1.3
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 10:55 Correction on Google's Open Source History
Ion reframes Sonal's assumption that Google was an open-source pioneer, pointing out that historical core infrastructure like MapReduce and GFS were closed source.
Hardest push from the host ▶ 18:55 Host Challenges Academic AI Lab RelevanceSonal directly pushes back against the guest, questioning why an academic lab is even needed when corporate AI labs have superior talent and funding.
Biggest teaching moment ▶ 16:40 Explaining System Robustness via the Elephant ExampleIon clearly educates the hosts on AI edge cases, demonstrating how a model trained on cats and dogs must gracefully handle unseen inputs like an elephant.
The host holds their own ▶ 23:35 Peter's Thesis on Data-Driven Computing vs Traditional CodingPeter demonstrates extensive domain expertise by explaining how traditional if-then-else coding logic has peaked, making data inputs the core driver of future computing.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Balancing Academia with Commercial Entrepreneurship | 4 | 5 | 0 | 0 | Peter Levine articulates how big data projects like Mesos expanded beyond their original scope into general resource orchestration. Ion explains UC Berkeley's institutional structure, including the strict 5-year cap on labs. | |
| Bridging Academic Papers and Real-World Industry Deployment | 4 | 5 | 0 | 1 | Sonal Chokshi highlights the friction between open-ended academic papers and production-scale industrial deployment. Ion explains how PhD students bridge this gap through internships and close production partnerships with companies like Twitter and Facebook. | |
| Open Source as the Engine of Global Software Innovation | 5 | 6 | 2 | 1 | Peter delivers a detailed overview of open source commercialization models like SaaS and cloud hosting. Ion gently corrects Sonal on Google's historical stance toward open source, clarifying that Google was mostly closed-source until recently. | |
| Community Building through Retreats, Camps, and Summits | 3 | 7 | 0 | 0 | Ion dominates the segment explaining the transition from big data analytics to real-time decision making in RISELab. He provides detailed pedagogical explanations of system requirements like robustness and explainability. | |
| Academic Research Relevance in the Era of Corporate AI Labs | 6 | 5 | 2 | 7 | Sonal directly pushes back on the premise of the new lab, questioning how academic labs remain relevant when corporate AI labs like Baidu and OpenAI have massive capital. Ion defends academic labs by pointing out corporate desires for neutral open-source stacks. | |
| Edge Computing, Data-Driven Execution, and Program Synthesis | 6 | 4 | 0 | 0 | Peter lays out his high-level thesis regarding edge computing and the end of traditional programming logic in favor of data-driven execution. Ion agrees and ties Peter's thesis directly to reinforcement learning and program synthesis. | |
| Key RISELab Projects and the Rules of Lab Naming | 4 | 5 | 0 | 0 | Ion details key RISELab projects including Ray, Clipper, Opaque, and Ground. The conversation wraps up with a lighthearted discussion of David Patterson's acronym rules for lab naming. |
Statements from this episode (12)
Stoica: Silicon Valley proximity keeps UC Berkeley research anchored in industry problems
“Academia, it's allowing you to do more experimentation. It's set for that. And at Berkeley, we are in a privileged position. Of course, being close to the Silicon Valley, we have a lot of feedback. From the industry. So we are very anchored in what are the rea…”
Stoica: UC Berkeley computer science research labs operate on five-year limits
“So the labs are like one rule is around five years.”
Stoica: Apache Spark was created to prove Mesos's value
“So actually, Mesos was the first project we developed. This is what started the stack. And Mesos was by design to support multiple cluster computing framework. We started with Hadoop. And actually, one of the reasons we designed Spark, it's To show that it's m…”
Stoica: Facebook's early big data cluster had 80 nodes and three people
“When we started working with Facebook, Facebook, you know, has an entire cluster, big cluster for big data. It was 80 nodes. And their big data team was like three people.”
Stoica: UC Berkeley's AMPLab generates zero patents and commits to open source
“The other thing about what AMLAB and all these labs have done is that they take a very strong stand about being open source and we generate no patents.”
Levine: Cloud and SaaS enabled the commercial monetization of open source
“Well, the element, I believe, that has brought open source from fringe hobby project out of academia to mainstream is the fact that you can actually build commercial businesses On top of open source. Right. And a lot of that, which you guys at Databricks are d…”
Chokshi: Microsoft has more GitHub commits than any other company
“The most interesting thing I saw this past week is this stat that Microsoft now has more commits to GitHub than any other company. And can you just think about that for a moment? Like Microsoft on GitHub.”
Stoica: Over 60% of UC Berkeley PhD applicants apply for AI
“When you look at the PhD students who applied to Berkeley, PhD applicants. It turns out that well over 60% are applying for AI.”
Stoica: Many published AI research algorithms are hard to reproduce
“Many of the algorithms which are published are hard to reproduce.”
Stoica: Computing functionality will migrate bidirectionally between cloud and edge
“Things which are now done in the cloud is going to migrate some of the functionality for on the edge. Also, on, in, on the other side, things which are now done only at the edge, like self-driving cars will migrate, some of the functionality will migrate to th…”
Levine: Traditional imperative logic programming is reaching the end of its lifecycle
“I believe that programming is sort of in the life cycle of Programming is sort of at the end. Like, you can only write so many if-then-else statements. You're kind of done with logic, right, in that sense.”
Stoica: ML models degrade over time as real-world data evolves
“The models you developed on some data set, and for instance, the data or the queries are going to evolve over time. And because the environment or the world around you evolves, what you learned and which is embedded in the model may not be as relevant or as go…”