{
  "version": "1.2.0",
  "chapters": [
    {
      "title": "Cold open\r",
      "startTime": 0,
      "endTime": 77
    },
    {
      "title": "Who’s Beth Barnes?\r",
      "startTime": 77,
      "endTime": 111
    },
    {
      "title": "Can we see AI scheming in the chain of thought?\r",
      "startTime": 111,
      "endTime": 556
    },
    {
      "title": "The chain of thought is essential for safety checking\r",
      "startTime": 556,
      "endTime": 770
    },
    {
      "title": "Alignment faking in large language models\r",
      "startTime": 770,
      "endTime": 1053
    },
    {
      "title": "We have to test model honesty even before they're used inside AI companies\r",
      "startTime": 1053,
      "endTime": 1622
    },
    {
      "title": "We have to test models when unruly & unconstrained\r",
      "startTime": 1622,
      "endTime": 1916
    },
    {
      "title": "Each 7 months models can do tasks twice as long\r",
      "startTime": 1916,
      "endTime": 3091
    },
    {
      "title": "METR's research finds AIs are solid at AI research already\r",
      "startTime": 3091,
      "endTime": 3498
    },
    {
      "title": "AI may turn out to be strong at novel & creative research\r",
      "startTime": 3498,
      "endTime": 3704
    },
    {
      "title": "When can we expect an algorithmic 'intelligence explosion'?\r",
      "startTime": 3704,
      "endTime": 4075
    },
    {
      "title": "Recursively self-improving AI might even be here in 2 years — which is alarming\r",
      "startTime": 4075,
      "endTime": 4469
    },
    {
      "title": "Could evaluations backfire by increasing AI hype & racing?\r",
      "startTime": 4469,
      "endTime": 5452
    },
    {
      "title": "Governments first ignore new risks, but can overreact once they arrive\r",
      "startTime": 5452,
      "endTime": 5995
    },
    {
      "title": "Do we need external auditors doing AI safety tests, not just the companies themselves?\r",
      "startTime": 5995,
      "endTime": 6849
    },
    {
      "title": "A case against safety-focused people working at frontier AI companies\r",
      "startTime": 6849,
      "endTime": 7720
    },
    {
      "title": "The new, more dire situation has forced changes to METR's strategy\r",
      "startTime": 7720,
      "endTime": 8215
    },
    {
      "title": "AI companies are being locally reasonable, but globally reckless\r",
      "startTime": 8215,
      "endTime": 8509
    },
    {
      "title": "Overrated: Interpretability research\r",
      "startTime": 8509,
      "endTime": 8624
    },
    {
      "title": "Underrated: Developing more narrow AIs\r",
      "startTime": 8624,
      "endTime": 9028
    },
    {
      "title": "Underrated: Helping humans judge confusing model outputs\r",
      "startTime": 9028,
      "endTime": 9175
    },
    {
      "title": "Overrated: Major AI companies’ contributions to safety research\r",
      "startTime": 9175,
      "endTime": 9405
    },
    {
      "title": "Could we have a science of translating AI models' nonhuman language or neuralese?\r",
      "startTime": 9405,
      "endTime": 9555
    },
    {
      "title": "Could we ban using AI to enhance AI, or is that just naive?\r",
      "startTime": 9555,
      "endTime": 9931
    },
    {
      "title": "Open-weighting models is often good & Beth has changed her attitude to it\r",
      "startTime": 9931,
      "endTime": 10222
    },
    {
      "title": "What we can learn about AGI from the nuclear arms race\r",
      "startTime": 10222,
      "endTime": 11153
    },
    {
      "title": "Infosec is so bad that no models are truly closed-weight models\r",
      "startTime": 11153,
      "endTime": 11443
    },
    {
      "title": "AI is more like bioweapons because it undermines the leading power\r",
      "startTime": 11443,
      "endTime": 12072
    },
    {
      "title": "What METR can do best that others can't\r",
      "startTime": 12072,
      "endTime": 13011
    },
    {
      "title": "What METR isn't doing that other people have to step up and do\r",
      "startTime": 13011,
      "endTime": 13329
    },
    {
      "title": "Research METR plans to do next",
      "startTime": 13329,
      "endTime": 0
    }
  ]
}