Sep 10, 2024 · 20m · allin
Sergey Brin | All-In Summit 2024
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
In an unannounced appearance at the All-In Summit 2024, Google co-founder Sergey Brin discusses his active return to hands-on AI development at Google, the architectural evolution of large language models, and the economic and competitive dynamics shaping the future of technology.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 3.9 Guest teaching 4.1 Guest disagreement 1.7 The hosts pushing back 2.6
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 8:13 Pushing back on extreme compute scaling projections
Sergey directly rejects widespread industry articles extrapolating multi-gigawatt compute demands, arguing algorithmic gains outpace raw hardware expansion.
Hardest push from the hosts ▶ 14:58 Confronting corporate conservatism at GoogleChamath challenges Sergey directly with an internal story about Google engineers hesitating to push AI features due to fear of error.
Biggest teaching moment ▶ 6:57 Explaining multi-model architectures vs God modelsSergey breaks down how Google used three distinct specialized models for the International Math Olympiad and how those lessons are integrated into unified models.
The host holds their own ▶ 14:58 Citing internal Google product culture anecdotesChamath displays insider reporting by bringing up a specific unreleased anecdote about Sergey encouraging engineers to ship AI coding tools.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Introductory Video Package Profiling Sergey Brin | 2 | 3 | 1 | 1 | Chamath opens with broad introductory questions regarding Sergey's return to active development at Google. Sergey humbly downplays his role and shares personal anecdotes about using AI for coding Sudoku puzzles. | |
| Unified AI Architectures Versus God Models | 5 | 5 | 2 | 2 | Chamath demonstrates solid domain awareness by citing Google's published paper on graph neural networks. Sergey clarifies how specialized AI architectures in the Math Olympiad are being merged back into unified general language models rather than single 'god models'. | |
| Compute Scaling, Energy Demand, and Market Rationality | 4 | 4 | 3 | 3 | Chamath pushes on whether current hardware infrastructure buildout is rational. Sergey expresses skepticism toward extreme gigawatt-scale extrapolations, noting algorithmic efficiency gains outpace raw compute increases. | |
| AI Applications in Biology, Robotics, and Lessons Learned | 4 | 4 | 1 | 3 | Chamath highlights Google's past record of selling or spinning out robotics ventures. Sergey self-deprecatingly admits Google started multiple robotics divisions too early before modern multimodal LLM foundations existed. | |
| Product Vision, Project Astra, and Production Challenges | 3 | 4 | 1 | 2 | Chamath asks about product vision and user interaction paradigms. Sergey explains the significant engineering gap between a 90% accurate impressive demo like Project Astra and a robust production product. | |
| Overcoming Internal Conservatism and Embracing AI Risk | 5 | 5 | 2 | 4 | Chamath brings up a specific insider story regarding Sergey overriding Google's internal conservatism around AI coding tools. Sergey candidly agrees that Google was initially too timid with transformers due to fear of public mistakes. | |
| Global AI Competition, Early Internet Parallels, and Conclusion | 4 | 4 | 2 | 3 | Chamath frames global AI competition versus total value creation. Sergey notes friendly competitive fire while pointing out Gemini recently topped the LMSYS ELO leaderboard, drawing parallels to the early web era. |
Statements from this episode (13)
Sergey Brin confirms he works at Google 'pretty much every day'
“Honestly, like, pretty much every day.”
Sergey Brin: Google engineers underutilize AI tools for their own coding
“They don't honestly use the AI tools for their own coding as much as I think they ought to.”
Brin: Google AI achieved a silver medal score at the Math Olympiad
“Like the International Math Olympiad that we participate in, we got silver medal as an AI, actually one point away from gold.”
Sergey Brin: AI development is trending toward unified, shared architectures
“But I do think the trend is to have a more unified model. I don't know if I'd call it a god model. But to have certainly sort of shared architectures and ultimately even shared models.”
Sergey Brin: AI algorithmic improvements may be outpacing compute scaling
“The algorithmic improvements that have come over the course of the last few years, ah, maybe are actually even outpacing the increased compute that's put into these models.”
Sergey Brin: Google Cloud turns away customers due to compute shortages
“For us, we're kind of building out compute as quickly as we can, and we just have a huge amount of demand. I mean, for example, our cloud customers just want a huge amount of TPUs, GPUs, you name it. You know, we just can't, we have to turn down customers beca…”
Sergey Brin: Language-model robotics are impressive but lack day-to-day robustness
“Robotics, For the most part, I see in this sort of wow stage, like wow, you could make a robot do that with just, you know, this general purpose language model, or just a little bit of fine tuning this way or that, and it's like amazing but maybe not for the m…”
Sergey Brin: Pre-multimodal robotics efforts feel 'silly' in hindsight
“It, yeah, it just feels kind of silly having done all of that work and seeing now how capable these general language models are that include, for example, vision and image, and they're multimodal, and they can understand The scene and everything, and not havin…”
Sergey Brin: Rapid base AI progress makes long-term product forecasting difficult
“It's, like, just really hard To, you know, just forecast, like, you know, to think five years out, because, you know, based on the base technical capability of the AI is what enables the applications and then sometimes, you know, somebody will just whip up a l…”
Sergey Brin: Moving 90% accurate AI demos to production requires massive effort
“It does it correctly, like, 90% of the time, but am I really, like, is that then worth it if 10% of the time it's gonna make a mistake or taking too long or whatever? And then you have to work, work, work, work, work, work, work to get to perfect all those thi…”
Sergey Brin admits Google was 'too timid' to deploy language models
“We were too timid to deploy them”