Skip to content
Now

NowThe sources

Every figure has someone who measured it. And it also says what it does not measure.

On this website a figure is always a measure of something that already happened. What does not exist yet is told in words.

  1. METR · May 2024 and April 2026

    Over 16 h, from 7 min.

    Its data file gives 17.4 h for the best model of April 2026, and METR warns that above 16 h its measure is unreliable. An earlier METR paper, from 2025, gave that same 2024 model 9 min on its task suite of the time: here both ends come from one suite. Doubling the length does not double what gets automated.
  2. DX · first and second quarter of 2026

    51.9%, from 27.4%.

    It is the average of a panel of more than 400 companies, not a measure of all. The figure is DX’s; we read it at developertoolkit.ai, which cites it.
  3. OpenAI · October 2026

    722 papers. The company’s own claim.

    3 were retracted and 14 were revised within days. The problem open since 1946 was solved in May 2026.
  4. McKinsey, The State of AI, 2026

    37% attribute something to it. 63%, nothing.

    It is a survey: 1,719 people in 97 countries.
  5. Menlo Ventures · 3 November 2026

    $176.8 B a year on using AI.

    It is a yearly rate as of September, not a year’s total. And it is an investor’s estimate, from a survey of US companies.
  6. The rules

    No source, no entry. From before 2026, it does not count. For what does not exist yet, none.

    Each figure has its row in a register, with the link and the exact sentence it comes from. What is coming is told in words.

Go on from here

StudioWhere we work. Each night is watched there while it runs.Keep going down