I edited to add some stuff about GWP and training compute for the most expensive model.
I agree that this focuses on qualitative stuff, but that’s only due to lack of good ideas for quantitative metrics worth tracking. I agree GWP and training compute are worth tracking, thank you for reminding me, I’ve edited to be more explicit.
I am not entirely sure why I didn’t think of the number of parameters as a high-level metric. Idk, maybe because it was weaved into the prose I didn’t notice it? My bad.
(To be clear, this wasn’t meant to be a critique, just a statement of what kind of forecast it was. I think it’s great to have forecasts of this form too.)
New planned summary:
This post describes the author’s median expectations around AI from now until 2026. It is part I of an attempt to write a detailed plausible future trajectory in chronological order, i.e. incrementally adding years to the story rather than writing a story with the end in mind. The hope is to produce a nice complement to the more abstract discussions about timelines and takeoff that usually occur. For example, there are discussions about how AI tools are used by nations for persuasion, propaganda and censorship.
I edited to add some stuff about GWP and training compute for the most expensive model.
I agree that this focuses on qualitative stuff, but that’s only due to lack of good ideas for quantitative metrics worth tracking. I agree GWP and training compute are worth tracking, thank you for reminding me, I’ve edited to be more explicit.
I am not entirely sure why I didn’t think of the number of parameters as a high-level metric. Idk, maybe because it was weaved into the prose I didn’t notice it? My bad.
(To be clear, this wasn’t meant to be a critique, just a statement of what kind of forecast it was. I think it’s great to have forecasts of this form too.)
New planned summary:
That’s great, thanks!