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

1kṀ3085

2035

10%

<2026

19%

<2027

32%

<2028

47%

<2029

50%

<2030

62%

<2035

38%

>=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.)

Technical AI Timelines

AI Safety

Get

1,000

to start trading!

People are also trading

Will an AI model use more than 1e28 FLOPS in training before 2026?

8% chance

Will the largest machine learning training run (in FLOP) as of the end of 2025 be in the United States?

86% chance

Will a machine learning training run exceed 10^26 FLOP in China before 2026?

52% chance

Will there be an announcement of a model with a training compute of over 1e30 FLOPs by the end of 2025?

5% chance

By 2030 will China have reached AI flops/$ parity with the US?

37% chance

Will open-source AI win (through 2025)?

31% chance

Will the largest machine learning training run (in FLOP) as of the end of 2040 be in the United States?

46% chance

Will the largest machine learning training run (in FLOP) as of the end of 2035 be in the United States?

69% chance

Will the largest machine learning training run (in FLOP) as of the end of 2030 be in the United States?

74% chance

Which of the following will be true about the next major AI breakthrough? (after test-time compute)

6 Comments

32 Holders

80 Trades

Sort by:

It is very likely that neither US government nor many companies will publish their FLOPs. "Open"AI for example is pretty secretive about metrics such as parameter count. Do we know how many FLOPS they use right now?

@ProjectVictory Yes this is a real issue, but we'll count https://epoch.ai/ estimates and other 3rd party analyses from teams of researchers. In general, GPU sales aren't too secret and I expect that to continue being the case.

Epoch AI

Epoch AI is a research institute investigating key trends and questions that will shape the trajectory and governance of Artificial Intelligence.

bought Ṁ100 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.