Key ideas from Julian Schrittwieser

• "We are seeing this very consistent improvement over many many years where every say like you know 3 4 months is able to like do a task that is twice as long as before completely on its own" - Julian
• By mid-2026, Julian extrapolates agents will work autonomously for a full day; by late 2026, at least one model matches industry experts across many occupations
• "By 2027 2028, I think extremely likely that the models will be smart enough and capable enough to actually have that level of insight" for Nobel Prize-level discoveries - Julian
• Move 37 from AlphaGo demonstrated AI creativity in 2016, playing an unexpected move that surprised professional Go players and ultimately won the game
• "Pre-training on this vast data sets we have just brings us so much value that we would from a practical point of view not want to give it up" - Julian on combining pre-training with RL
• "The main question is do these two trends balance each other out so that you know the AI makes us increasingly more productive" versus problems getting harder - Julian on AI discontinuity
• Reinforcement learning training is much more unstable than supervised learning due to feedback cycles, requiring careful isolation of components during development
• "If you manage to make everybody in society 10 times more productive you know what kind of abundance can we achieve?" - Julian on AI's economic potential