{
  "version": "1.2.0",
  "chapters": [
    {
      "title": "Cold open\r",
      "startTime": 0,
      "endTime": 48
    },
    {
      "title": "Who's Allan Dafoe?\r",
      "startTime": 48,
      "endTime": 87
    },
    {
      "title": "Allan's role at DeepMind\r",
      "startTime": 87,
      "endTime": 267
    },
    {
      "title": "Why join DeepMind over everyone else?\r",
      "startTime": 267,
      "endTime": 557
    },
    {
      "title": "Do humans control technological change?\r",
      "startTime": 557,
      "endTime": 1224
    },
    {
      "title": "Arguments for technological determinism\r",
      "startTime": 1224,
      "endTime": 1589
    },
    {
      "title": "The synthesis of agency with tech determinism\r",
      "startTime": 1589,
      "endTime": 2233
    },
    {
      "title": "Competition took away Japan's choice\r",
      "startTime": 2233,
      "endTime": 2529
    },
    {
      "title": "Can speeding up one tech redirect history?\r",
      "startTime": 2529,
      "endTime": 2875
    },
    {
      "title": "Structural pushback against alignment efforts\r",
      "startTime": 2875,
      "endTime": 3145
    },
    {
      "title": "Do AIs need to be 'cooperatively skilled'?\r",
      "startTime": 3145,
      "endTime": 3719
    },
    {
      "title": "How AI could boost cooperation between people and states\r",
      "startTime": 3719,
      "endTime": 4018
    },
    {
      "title": "The super-cooperative AGI hypothesis and backdoor risks\r",
      "startTime": 4018,
      "endTime": 4402
    },
    {
      "title": "Aren’t today’s models already very cooperative?\r",
      "startTime": 4402,
      "endTime": 4582
    },
    {
      "title": "How would we make AIs cooperative anyway?\r",
      "startTime": 4582,
      "endTime": 4944
    },
    {
      "title": "Ways making AI more cooperative could backfire\r",
      "startTime": 4944,
      "endTime": 5416
    },
    {
      "title": "AGI is an essential idea we should define well\r",
      "startTime": 5416,
      "endTime": 6061
    },
    {
      "title": "It matters what AGI learns first vs last\r",
      "startTime": 6061,
      "endTime": 6339
    },
    {
      "title": "How Google tests for dangerous capabilities\r",
      "startTime": 6339,
      "endTime": 7066
    },
    {
      "title": "Evals 'in the wild'\r",
      "startTime": 7066,
      "endTime": 7304
    },
    {
      "title": "What to do given no single approach works that well\r",
      "startTime": 7304,
      "endTime": 7534
    },
    {
      "title": "We don't, but could, forecast AI capabilities\r",
      "startTime": 7534,
      "endTime": 7885
    },
    {
      "title": "DeepMind's strategy for ensuring its frontier models don't cause harm\r",
      "startTime": 7885,
      "endTime": 8101
    },
    {
      "title": "How 'structural risks' can force everyone into a worse world\r",
      "startTime": 8101,
      "endTime": 8375
    },
    {
      "title": "Is AI being built democratically? Should it?\r",
      "startTime": 8375,
      "endTime": 8674
    },
    {
      "title": "How much do AI companies really want external regulation?\r",
      "startTime": 8674,
      "endTime": 9201
    },
    {
      "title": "Social science can contribute a lot here\r",
      "startTime": 9201,
      "endTime": 9355
    },
    {
      "title": "How AI could make life way better: self-driving cars, medicine, education, and sustainability",
      "startTime": 9355,
      "endTime": 0
    }
  ]
}