Feb 18, 2026 · 47m · big-technology
How Google DeepMind Operates & Experiments — With Lila Ibrahim and James Manyika
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
Recorded live at the World Economic Forum in Davos, Google executives Lila Ibrahim and James Manyika explain how Google DeepMind and Google Labs balance rigorous scientific research with rapid product innovation. They detail the development of flagship tools like Gemini and NotebookLM while showcasing groundbreaking frontier bets across quantum computing, material discovery, climate prediction, and space-based compute.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 27.3% of the talking time here. How this is scored →
Alex as informed peer 4.7 Guest teaching 4.6 Guest disagreement 1.0 Alex pushing back 2.8
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:47 James reframes Google's shipping record
James directly addresses Sam Altman's claim that Google was afraid to ship, arguing that balancing bold innovation with safety is an essential, deliberate tension rather than institutional paralysis.
Hardest push from Alex ▶ 25:05 Alex quotes Sam Altman on Google's lost leadAlex challenges the corporate narrative by quoting Sam Altman's assertion that Google could have crushed OpenAI early if it had taken them seriously and shipped faster.
Biggest teaching moment ▶ 38:20 James details the breakthrough in quantum error correctionJames gives an in-depth explanation of below-threshold error correction on the Willow chip, showing how error rates decrease as qubits scale up.
Alex holds their own ▶ 46:07 Alex cites Ilya Sutskever on planetary compute limitsAlex contextualizes Google's space compute project by referencing Ilya Sutskever's conceptual argument regarding data center sprawl on Earth.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Google DeepMind's Operating Model and Philosophy | 4 | 5 | 1 | 2 | Alex asks structured questions about DeepMind's operating model, contrasting it with OpenAI's startup-cluster structure. Lila explains the blend of top-down ambition from Demis Hassabis with bottom-up interdisciplinary exploration. | |
| Merging Research with Products: The Gemini Engine Room | 5 | 4 | 2 | 4 | Alex advances his thesis regarding Sundar Pichai's McKinsey background driving centralization into a single AI 'engine room.' Lila softly rejects the term 'farming out' research, while James explains the unification of Google Brain and DeepMind. | |
| Rebooting Google Labs: NotebookLM and AI-First Products | 5 | 3 | 0 | 2 | A collaborative segment where James walks through the genesis of NotebookLM and Audio Overviews. Alex demonstrates deep familiarity with the products, sharing an anecdote about using Google Flow for his mountain climb video. | |
| Cultivating Innovation: 20% Time and Research-to-Reality | 5 | 3 | 1 | 2 | Alex presses on whether Google's famous 20% time culture actually still exists, citing historical S&P 500 company longevity. James and Lila confirm that 20% exploration continues to drive internal projects like Learn Your Way and Project Aeneas. | |
| Balancing Responsibility with the Urgency to Ship | 6 | 4 | 3 | 6 | Alex directly confronts the guests with Sam Altman's quote that Google was afraid to ship and could have smashed OpenAI early on. James pushes back by framing the hesitation as a necessary tension between speed and responsible deployment. | |
| Transforming Education with AI and LearnLM | 5 | 5 | 1 | 3 | Alex presents user survey stats and points out real-world issues like cognitive offloading and professor misuse reported by the NYT. Lila and James elaborate on pedagogy-first design with LearnLM and guided learning frameworks. | |
| Frontier AI Bet 1: Quantum Computing Milestones | 3 | 7 | 0 | 1 | Alex invites an update on quantum computing, allowing James to conduct a detailed masterclass on superconducting qubits, below-threshold error correction on the Willow chip, and useful NMR spin dynamic calculations. | |
| Frontier AI Bet 2: Material Science with GNoME | 4 | 5 | 1 | 3 | Alex asks about material discovery through GNoME and actively manages time constraints while discussing AI weather forecasting models like GraphCast and riverine flood predictions across 150 countries. | |
| Frontier AI Bet 4: Project Suncatcher and Space Computing | 5 | 5 | 0 | 2 | James reveals Project Suncatcher's timeline to put TPUs in space by 2027. Alex connects this to Ilya Sutskever's theory on overcoming data center land constraints before wrapping the episode. |
Statements from this episode (16)
Ibrahim: Interdisciplinary teams pairing ethicists and scientists unlock breakthrough AI research
“We want to build interdisciplinary teams. How do you create a culture where you can have a bioethicist next to a computer scientist and a neuroscientist? Because we think that's really where the magic happens and unlocks the work.”
Ibrahim: Hassabis greenlit infusing learning science into Gemini when model matured
“Over the past few years, we've been doing a lot of work around one science area, learning science. How do people learn and can we improve it? And then this year was really, Dennis was like, okay, Gemini is good enough. It's time to infuse Everything we've done…”
Manyika: Google DeepMind develops specialized AI models for Waymo's autonomous driver
“And then you've got all these other, you know, special kind of ambitious projects working on things like Genie build world models. You've got work going on to build special things for Waymo and enhance the models that Waymo's that lead to Waymo, the Waymo driv…”
Manyika: NotebookLM originated as Project Tailwind with a five-person team
“So it started out as a protocol tailwind. There were four, five people working on it.”
Manyika: Google Labs actively runs about 30 different experimental projects
“So if you go to the Google lab site, you'll probably see about 30 different things”
Manyika: 20% of Google Labs projects stem from employee 20% time
“So if you think about the things that are in labs, I would say something like maybe. 80% of them came out of people actually in the labs team. The other 20% came from 20% stuff.”
Manyika: 3.5 Million Researchers Across 190 Countries Access AlphaFold
“You now have three and a half million researchers accessing it in over a 190 countries.”
Manyika: Google releases new Gemini model generations every five to six months
“We now kind of on the cycle that Gemini models where every five, six months, there's the latest generation.”
Ibrahim: Northern Ireland AI pilot saved teachers 10 hours weekly
“What they found was on average they were sending 10
Saving 10 hours per week per teacher, and their post-it notes were how they were using their time, which was, I'm getting time back with my family. I can now do lesson plans for different learners of differen…”
Manyika: Google Willow chip solved 10-septillion-year benchmark in under five minutes
“Our Willow chip, for example, hit a big milestone. It was a big enough deal about a year, year and a half ago where, you know, it was able to do in, you know, a computation, a benchmark computation called RCS, which would take a classic frontier supercomputer,…”
Manyika: Google demonstrated below-threshold quantum error correction
“For the first time, we're able to show that you can do what's called below threshold error correction, which is as you scale up the system, the error rates are actually going down, which is exactly what you'd want, as opposed to that they're actually going up.”
Manyika: Google demonstrated first useful quantum computation for molecular spin dynamics
“Last year we were able to show probably the first useful computation. These are quantum echoes results. It was a big enough deal made the cover of nature, which is great. Our teams are excited about that. What that showed was an actual useful computation for f…”
Manyika: Useful quantum computing applications will emerge within five years
“So I actually think we're going to start to see useful applications in the next five or so years from quantum computing.”
Ibrahim: DeepMind's GNoME expanded known stable crystals from 40,000 to 400,000+
“The alpha genome, the alpha gnome, the one that you've just mentioned, our material science, was really exciting because we basically went from 40,000 known stable crystals to 400,000 plus. That are now being tested with research, in research and in labs.”