The case for intelligence beyond language

<p>Raia Hadsell, VP of Research at Google DeepMind, makes the case that intelligence is more than language: the same recipe that learns the patterns of text can learn the patterns of any complex system. She walks through DiffusionGemma and text diffusion, the Genie world models, and DeepMind&#39;s robotics stack, where world models now generate training data you cannot tell from the real thing. Recorded at RAAIS 2026.</p><p><strong>Timestamps</strong></p><p>0:00 Introduction (Nathan Benaich)</p><p>0:35 From philosophy to DeepMind: the frontiers of intelligence</p><p>2:37 The twenty-year lesson: one recipe for complex systems</p><p>4:34 DiffusionGemma and the Gemma 4 open models</p><p>5:32 How text diffusion works</p><p>7:34 Speed, self-correction, and the sudoku test</p><p>10:51 World models: better agents need better worlds</p><p>12:56 Genie 1 to Genie 3</p><p>15:01 Genie 3 demos: typing a world into being</p><p>18:20 World models for education</p><p>19:38 Grounding Genie in Street View</p><p>20:41 Robotics: a brain and a spine</p><p>23:14 Gemini Robotics-ER 1.6 and Boston Dynamics&#39; Spot</p><p>24:42 The vision-language-action model</p><p>25:45 The data bottleneck and closing the loop</p><p>27:24 Beyond language: the domains still to crack</p><p><br></p>