From driving the world to dreaming it

<p>World models are the bet that AI should learn the world by watching it and acting in it, not just by reading about it. At RAAIS 2026, Odyssey co-founder and CTO Jeff Hawke makes the case: a world model is a neural simulator - an interactive stream of pixels that runs in real time, models physics, and answers back.</p><p>He walks through Odyssey&#39;s four research fronts - streaming interactive pixels (Odyssey-2), joint audio and video (Starchild-1), shared multiplayer worlds (Agora-1, demoed live as a fully generated game of GoldenEye), and PROWL, which sends a reinforcement-learning agent to find and fix a world model&#39;s own failures - and argues the field is at its GPT-2 moment: promising, but pre-ChatGPT, with the GPT-3-style commercial unlock still ahead.</p><p>Recorded at the 10th Research and Applied AI Summit (RAAIS), London, June 2026.</p><p>Timestamps</p><p>00:00 Intro: Nathan on Odyssey and world models</p><p>01:05 Jeff Hawke: from self-driving to world models</p><p>01:40 The bet — a missing form of intelligence</p><p>02:40 Why world models suddenly matter (the late-2025 flip)</p><p>03:16 What a world model actually is (and isn&#39;t)</p><p>04:45 The neural simulator</p><p>06:34 Two principles: end-to-end learning and generality</p><p>07:19 The &quot;GPT-3 of world models&quot; and four research themes</p><p>08:46 Odyssey-2: streaming, interactive pixels</p><p>10:33 Starchild-1: generating audio and video together</p><p>13:03 Agora-1: multiplayer world models</p><p>13:57 Live demo: the room plays GoldenEye</p><p>16:20 PROWL: improving the model by breaking it</p><p>18:39 Where Odyssey goes next</p><p>19:55 Still the GPT-2 era</p><p>21:30 Q&amp;A: physics limits, safety, compute cost, merging with LLMs</p>