When will a US government AI run overtake private AI compute by FLOP?

Plus

Ṁ2971

2035

1.9%

<2026

19%

<2027

27%

<2028

35%

<2029

50%

<2030

66%

<2035

50%

>=2035

By what year will the most FLOP intensive run have been conducted or owned by the US federal government? Resolves Yes on all later years.

Qualifying events:

Any arrangement (contractors, national labs etc.) under which the US government employees have the right to control what queries are run on the model.
Any international collaboration where the US retains model query veto rights, but does not include international collaboration where the US is a stakeholder having e.g. a minority vote.
If the US government seizes control over the current most FLOP-intensive pre-existing AI this qualifies.
Any unanticipated event which clearly satisfies the spirit of this question

Government-controlled AI runs will be compared to the most FLOP-intensive AI run by a US-registered company i.e. we are excluding non-US runs.

(In case this happened long ago incidentally at some national lab, we'll only count post-2023 runs.)

This question is managed and resolved by Manifold.

#AI

#Technical AI Timelines

#AI Safety

Get

1,000

and

3.00

4 Comments

29 Holders

65 Trades

Sort by:

bought Ṁ100 >=2035 NO

"Resolves Yes on all later years."

I assume this doesn't apply to the >=2035 bin. Like, if it happens in 2029, the <2035 resolves YES but >=2035 resolves NO?

@MaxHarms Correct

How does this resolve if the US govt takes control of a pre-existing model?

@DavidRein Good point. I'd say if the pre-existing model holds the FLOP record, then that would be sufficient for resolving positively. Will write that in for now. Open to argument if anyone wants to chime in here.

Related questions

Related questions