An interview with Google DeepMind CEO Demis Hassabis on why he doesn't believe we will have AGI in 2025, putting Google's assistants in smart glasses, and more
Just a bunch of engineers and MBAs refusing to read their opponents. Tim Moss / @tdmoss : Really interesting. Much more realistic take on both timing and definition of AGI [embedded post] X: Reed Albergotti / @reedalbergotti : The next phase of the AI race is about inference and Google's light chips could give it an edge. https://www.semafor.com/...
Context & Ripple Effects
This interview sits in a continuing DeepMind discussion that has moved from its post-restructuring AGI strategy to a stated need for capabilities beyond scaling and AI agents. It pairs a more cautious near-term AGI view with a concrete route for bringing Google assistants into a new device category.
The coverage also frames inference as a competitive front, with observers pointing to Google’s light chips as a possible advantage. That makes the interview relevant not only to AGI forecasting, but to how Google turns model capability into widely available products.
First-order effects
- Hassabis’s rejection of a 2025 AGI outcome tempers near-term expectations around DeepMind’s research timetable while keeping AGI central to Google’s strategy.
- Google’s assistant plans for smart glasses place its AI distribution effort in a wearable interface, while the inference-chip discussion highlights a potential infrastructure advantage.
Second-order effects
- Rivals pursuing AI wearables will face pressure to match both assistant usefulness and device-level distribution, rather than compete solely on model announcements.
- If inference becomes the next focal point of the AI race, efficiency of chips and serving infrastructure could matter more directly to product rollout and operating economics.
Third-order effects
- The story points to a two-track AI competition: longer-horizon work toward AGI alongside nearer-term competition to embed assistants in devices and operate them efficiently at scale.
- If this pattern persists, leadership will be shaped less by a single AGI deadline than by the combination of model progress, inference infrastructure, and distribution channels.
The trend: AI competition is shifting from headline model milestones toward the integrated race to serve assistants efficiently through owned hardware and consumer interfaces.