May 21, 2025 · 29m · big-technology
DeepMind CEO Demis Hassabis + Google Co-Founder Sergey Brin: AGI by 2030?
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
Google DeepMind CEO Demis Hassabis and Google Co-Founder Sergey Brin join Alex Kantrowitz on the Big Technology Podcast to discuss the road to Artificial General Intelligence, examining test-time reasoning architectures, multimodal embodiment, and the algorithmic breakthroughs needed by 2030.
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 21.6% of the talking time here. How this is scored →
Alex as informed peer 4.4 Guest teaching 3.7 Guest disagreement 1.3 Alex pushing back 2.6
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:05 Brin calls out Hassabis for timeline sandbagging
When asked if AGI arrives before or after 2030, Brin predicts before while Hassabis says just after, prompting Brin to playfully tell Hassabis to stop sandbagging and deliver it next week.
Hardest push from Alex ▶ 13:49 Kantrowitz challenges Hassabis on intelligence explosion risksKantrowitz directly confronts Hassabis on whether self-improving algorithmic systems like Alpha Evolve are intended to trigger an uncontrollable intelligence explosion.
Biggest teaching moment ▶ 8:48 Hassabis educates on true definition of AGI vs human task parityHassabis reframes the popular understanding of AGI, explaining that true AGI is not just matching typical human labor but matching the universal capability of the human brain architecture across historic geniuses.
Alex holds their own ▶ 22:31 Kantrowitz drills into model collapse from recursive synthetic dataKantrowitz demonstrates strong technical command by pressing Hassabis on whether synthetic video ingestion degrades model quality during subsequent training cycles.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Scaling Frontier Models vs Algorithmic Advances | 4 | 3 | 1 | 2 | Kantrowitz opens by framing the industry debate around diminishing returns on scaling frontier models. Both Hassabis and Brin agree that scaling and algorithmic breakthroughs must work together, with Brin providing the historical comparison of n-body simulations. | |
| Data Center Demands and Test-Time Compute | 5 | 5 | 1 | 3 | Kantrowitz probes into data center requirements and references his prior conversations with Hassabis regarding reasoning and test-time compute. Hassabis and Brin illustrate the power of inference compute by citing the 600 ELO rating jump in AlphaGo. | |
| Parallel Reasoning, True Invention, and World Models | 4 | 6 | 2 | 3 | Kantrowitz challenges the relevance of the term AGI amid industry fatigue. Hassabis provides an extensive theoretical re-definition, distinguishing between broad human architectural capability and narrow task completion while highlighting the brittleness of current models. | |
| The Race to AGI and Safety Implications | 5 | 4 | 1 | 4 | Kantrowitz asks whether AGI is a winner-take-all scenario, asks about emotional capacity in AI, and presses Hassabis on whether Alpha Evolve is designed to cause an uncontrolled intelligence explosion. Hassabis calmly clarifies the limits of evolutionary self-improvement in open-world settings. | |
| Sergey Brin's Return to Google and Technical Role | 3 | 2 | 1 | 2 | Kantrowitz asks Brin about his motivation for returning to active engineering at Google and his daily routine. Brin explains his hands-on involvement in Gemini pre- and post-training and humorously remarks on working alongside DeepMind leadership. | |
| Multimodal AI Agents and the Future of Robotics | 5 | 4 | 1 | 2 | Kantrowitz contrasts Google's camera-first, visual agent approach with competitors' screen-based assistants. Hassabis explains that real-world grounding and robotic embodiment require multimodal perception from day one. | |
| Lessons from Google Glass and the Vision for Smart Glasses | 4 | 2 | 1 | 2 | Kantrowitz asks Brin what lessons Google Glass offers for modern smart glasses. Brin candidly reflects on past hardware supply chain missteps and timing issues, while Hassabis argues AI assistants are the natural killer app for the form factor. | |
| Video Generation, Model Collapse, and SynthID | 6 | 5 | 1 | 3 | Kantrowitz raises the risk of model collapse if generative video floods future training data pools. Hassabis details data filtering solutions, SynthID watermarking, and the deliberate use of curated synthetic data pioneered in AlphaFold. | |
| Rapid-Fire Questions: Web Future, AGI Timeline, and Job Interviews | 4 | 1 | 2 | 3 | In a rapid-fire sequence, Kantrowitz quizzes both guests on the 10-year web outlook, AGI timelines before or after 2030, and using AI in hiring interviews. Brin and Hassabis engage in playful banter over timeline expectations. | |
| Simulation Hypothesis and Concluding Remarks | 4 | 5 | 2 | 2 | Kantrowitz brings up Hassabis's cryptic social media post about nature and simulations. Hassabis clarifies his computational universe view grounded in information theory, while Brin outlines the logical paradox of infinite simulation stacks. |
Statements from this episode (27)
Hassabis: Reaching AGI may require one or two more new breakthroughs
“And I think to get all the way to something like AGI, I think may require one or two more new breakthroughs.”
Brin: Algorithmic advances have historically beaten computational advances
“As you plot it, the algorithmic advances have actually beaten out the computational advances, even with Moore's law.”
Brin: Algorithmic advances will probably outpace compute advances in AI
“If I had to guess, I would say the algorithmic advances are probably going to be Even more significant than the computational advances, but both of them are coming up now, so we're kind of getting the benefits of both.”
Hassabis: Test-time compute and inference will require vast data center capacity
“We're going to need a lot of data centers for serving, and also for inference time compute. Giving, you know, you saw DeepThink today, T.P.P. Pro DeepThink. The more time you give it, the better it will be, and certain tasks, very high value, very difficult ta…”
Hassabis: Test-time compute elevated AlphaZero from master level to superhuman
“We had versions of AlphaGo and AlphaZero with the thinking turned off, so it was just the model telling you its first idea. And, you know, it's not bad, it's maybe like master level, something like that, but then if you turn the thinking on, it's way beyond wo…”
Brin: Inference-time compute offsets 5,000x in training compute
“What they did with AlphaGo and AlphaZero, as you mentioned it showed, as I recall, something you would take 5000 times as much training to match what you were able to do with still a lot of training and the inference time compute that you were doing with Go.”
Hassabis: Tool use during test-time reasoning will be extremely powerful
“Especially if you think about, obviously, with an AI, during its thinking process, it can also use a bunch of tools, or even other AIs in, during that thinking process to improve what the final output is. So I think it's gonna be an incredibly powerful paradig…”
Hassabis: AI Capable of True Scientific Invention Is Coming
“You know, that's, I think we don't have systems yet that can do that type of creativity. I think they're coming and these types of paradigms might be helpful in that things like thinking and then probably many other things.”
Hassabis: DeepMind Chief Scientist Shane Legg Co-Invented the Term AGI
“Shane Legg, who's our chief scientist who was one of the people who invented the term 25 years back.”
Hassabis: True AGI Must Match the Entire Capability Range of Human Brain Architecture
“What I would call AGI, is really a more theoretical construct, which is what is the human brain as an architecture able to do, right? And that's, the human brain's an important reference point because it's the only evidence we have maybe in the universe that g…”
Hassabis: AGI Should Take Experts Months to Find Obvious Flaws, Not Minutes
“For something to be called AGI, it would need to be consistent, much more consistent across the board Than it is today. It should take, like, a couple of months for maybe a team of experts to find a hole in it, an obvious hole in it. Whereas, you know, today i…”
Brin: Multiple entities will reach AGI quickly following the first breakthrough
“I guess I would suppose that one company or country or entity will reach AGI first. Now it is a little bit of a, you know, kind of a spectrum. It's not like a completely precise thing, so it's conceivable that there will be more than one roughly in that range …”
Hassabis: First AGI systems must be safe to spawn secure derivative architectures
“Assuming there is one, you know, there probably will be some organizations that get there first, and I think it's important to, that those first systems are built reliably and safely, and I think after that, if that's the case, you know, we can imagine using t…”
Hassabis: AGI Must Understand Emotion but Emulating It May Be Undesirable
“I think it will need to understand emotion. I don't know if I think it will be a sort of almost a design decision if we wanted to mimic emotions. I think there's no, I don't see any reason why it couldn't in theory. But it might be different or we might, it mi…”
Brin: Computer scientists should not be retired and should work on AI
“Anybody who's a computer scientist should not be retired right now, should be working on AI.”
Brin: Google fully intends Gemini to be the first AGI
“We fully intend that Gemini will be the very first AGI, clarify that.”
Brin: I work daily on Gemini pre-training and post-training models
“I'm across the street you know, pretty much every day and they're just people who are working on the key Gemini text models, on the pre-training, on the post-training, mostly those I periodically delve into some of the multimodal work”
Hassabis: Software intelligence, not hardware, has always held robotics back
“I've always felt that the bottleneck in robotics isn't so much the hardware, although obviously there's many, many companies, and working on fantastic hardware, and we partner with a lot of them, But it's actually the software intelligence that I think is alwa…”
Hassabis: Gemini 2.5 and video algorithms will make robotics work
“I think we're in a really exciting moment now where finally with these latest versions, especially 2.5 Gemini and more things that we're going to bring in this kind of video technology and other things. I think we're going to have really exciting algorithms to…”
Brin: Google Glass struggled because I lacked supply chain expertise
“There's also just I just didn't know anything about consumer electronics supply chains really, and how hard it would be to build that, and have it be at a reasonable price point managing all the manufacturing and so forth.”
Hassabis: Universal assistant is the killer app for smart glasses
“I feel like the universal assistant is the killer app for smart glasses, and I think that's what's going to make it work, apart from the fact that it's all, the tech, the hardware technology has also moved on, and Improved a lot. Is this, I think, I feel like …”
Hassabis: DeepMind trained AlphaFold using 300,000 to 400,000 synthetic predictions
“Where there wasn't actually enough real experimental data to build the final alpha fold, so we had to build an earlier version that then predicted about a million protein structures, and then we selected, it had a confidence level on that, we selected the top …”
Hassabis: Web will transform into an agent-first architecture within years
“I do think the web, I think in nearer term, the web is going to change quite a lot if you think about an agent first web. Like, does it really need to, you know, it doesn't necessarily need to see renders and things like we do as humans using the web, so I thi…”
Brin: Artificial General Intelligence will arrive before 2030
“2030. Boy, you really kind of, ah, put it on that fine line. I'm gonna say before.”
Hassabis: Underlying Physics Is Information Theory in a Computational Universe
“I do think that ultimately underlying physics is information theory. So I do think we're in a computational universe, but it's not just a straightforward simulation.”
Brin: Humans Lack Capacity to Reason About Higher-Level Simulations
“I think that we're taking a very anthropocentric view, like when we say simulation in the sense that some kind of conscious being is running a simulation that we are then in, and that, that they have some kind of semblance of desire and consciousness that's si…”