7
Checking in on AI 2027 a year later (awful.systems)
submitted 1 month ago* (last edited 1 month ago) by to c/sneerclub@awful.systems
6 comments fedilink hide all child comments
 

Originally posted about in the stubsack but decided to make this its own post because what the hell, let’s see how everyone’s favourite scenario is doing!

There are two takes on it: the author’s own (their 2025 evaluation is here) and another one from this website called AI 2027 tracker (not run by the authors) pretty sure this one only evaluates the qualitative predictions and not the quantitive ones

Both of them show that we are behind AI 2027’s predicted pace. By the author estimate: 2025 was only at 58% and 2026 is 69% of progress. AI 2027 tracker shows a consistent decrease in accuracy over time (from 100% to 81% to 91% to 46%)

all 6 comments

sorted by: hot top controversial new old
[–] 7 points 1 month ago* (last edited 1 month ago) (3 children)

Didn't someone come out with AI 2040?

I think as soon as they published it (ie. AI 2027) the original authors started to backpedal on timelines.

  • permalink
  • fedilink
  • source
  • hideshow 3 child comments
    •
  • [–] [S] 4 points 1 month ago* (last edited 1 month ago) (2 children)

    The same authors did publish AI 2040 yes

    They started backpedalling a fair bit earlier (at the beginning of 2026) their % chance of AGI in 2027 has halved from the release of AI 2027 to day (from 40% down to 23%)

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 7 points 1 month ago* (last edited 1 month ago) (1 child)

    In just 12 months, AI timelines have grown 550%. At this rate of growth, timelines will grow 178500% as soon as 2030, and years with the number 1785 in them have historically involved Frenchmen flying across the English channel and people in Mesopotamia doing awesome word problems. For the security of the United States we cannot allow a balloon and/or algebra gap and we propose a new Manhattan Project to sit with our polycule in Berkeley trying to solve cubic equations and designing really cool balloons with lasers and streaming anime.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
    • [–] [S] 4 points 1 month ago* (last edited 1 month ago) (1 child)

    Looked at AI 2027 tracker again and the way they grade the predictions is weird 80% accuracy is somehow both confirmed accurate and only partially accurate? huh?? why?? they do provide a grading criteria and 80% is listed as "largely accurate", yet for some reason they don't put that on the grading? It's either confirmed or partially accurate

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •