So your claim is that the state of the art weather models are accurate at somewhere between 24h and 2 weeks (unclear based on the other sub threads) and continually improving. Based on this you extrapolate that given enough compute and sensors it would be possible to predict the weather with the same accuracy a month or more out. I think that’s a reasonable claim assuming that (1) the behavior of the system is deterministic and (2) the system behaves the same at both time scales.

Setting (1) aside, I claim that (2) is a wrong assumption. That weather exhibits chaotic behavior and that this likely puts an upper bound on prediction accuracy and the upper bound is less than 1 month.

The state of the art appears to be GraphCast [1] and FengWu [2]. These show promise out past 2 weeks when run against historical data. However, neither model is making actual weather predictions, and both are still in preprint (e.g. methodology has not been peer reviewed). This is super interesting and it’s possible my claim is incorrect, and that the upper bound is further out than the conventional 2 week limit.

[1]: https://arxiv.org/html/2504.20238 [2]: https://arxiv.org/abs/2304.02948