This is a probably no situation: The estimated frontier threshold (10²⁶-10²⁷) is being discussed, so it’s plausible we’ll cross it — and before mid-2027 is possible given current trajectories.
No Scores Yet
Relative Brier Score
21
Forecasts
1
Upvotes
Forecasting Activity
Forecasting Calendar
| Past Week | Past Month | Past Year | This Season | All Time | |
|---|---|---|---|---|---|
| Forecasts | 0 | 2 | 22 | 21 | 37 |
| Comments | 0 | 2 | 13 | 13 | 20 |
| Questions Forecasted | 0 | 2 | 7 | 6 | 14 |
| Upvotes on Comments By This User | 0 | 0 | 1 | 1 | 4 |
| Definitions | |||||
New Prediction
New Prediction
Probability
Answer
25%
(-15%)
U.S. Entity
60%
(0%)
Non-U.S. Entity
Why do you think you're right?
Files
Why might you be wrong?
“U.S. or non-U.S. entity” is broad — so it simply means someone, somewhere, not necessarily a U.S. lab. That raises the probability somewhat, since it doesn’t restrict geo.
Files
New Prediction
Probability
Answer
40%
(0%)
U.S. Entity
60%
(0%)
Non-U.S. Entity
Confirmed previous forecast
Files
New Prediction
Probability
Answer
14%
(0%)
Yes
86%
(0%)
No
Confirmed previous forecast
Files
New Prediction
Probability
Answer
27%
(0%)
Less than 30 days
0%
(0%)
30 days
9%
(0%)
31-60 days
0%
(0%)
61-90 days
64%
(0%)
91 days or more
Confirmed previous forecast
Files
New Prediction
Probability
Answer
4%
(0%)
Estonia
3%
(0%)
Latvia
1%
(0%)
Lithuania
Confirmed previous forecast
Files
New Prediction
Probability
Answer
40%
U.S. Entity
60%
Non-U.S. Entity
Why do you think you're right?
Estimating a probability for U.S. entities crossing ≥ 10²⁷ FLOPs by the deadline (by 31 May 2027)
Given the info, I’d frame a subjective probabilistic estimate along these lines:
• Given that the top confirmed run is ~5 × 10²⁶ FLOPs, there is a gap (a factor of ~2) before hitting the 10²⁷ threshold.
• But the trend is sharply upward, and many models may be in “stealth” (i.e. unpublished compute) or underreported.
• The U.S. labs (OpenAI, xAI, etc.) have strong incentives, resources, and momentum to push the frontier.
So balancing the evidence, I’d put the probability for a U.S.-based entity having trained a model ≥ 10²⁷ FLOPs by May 2027 at somewhere in the 30 % to 50 % range, leaning toward the lower side given no public confirmation yet.
Files
Why might you be wrong?
Joint Probability
Since U.S. and non-U.S. probabilities are not mutually exclusive, we don’t add them — it’s possible both will cross ≥ 10²⁷ FLOPs by the deadline.
• Probability at least one (U.S. or non-U.S.) crosses ≥ 10²⁷ FLOPs by May 2027:
• Using independence approximation:
P(\text{≥10²⁷ by any}) ≈ 1 - (1 - 0.40)(1 - 0.20) ≈ 52 %
So there’s roughly a 50–50 chance globally that we see a model trained at or above 10²⁷ FLOPs by mid-2027.
Files
New Prediction
Probability
Answer
14%
(0%)
Yes
86%
(0%)
No
Confirmed previous forecast
Files
Why do you think you're right?
legislation is slowing us down ATM
Why might you be wrong?
AI is developing quite rapidly