Two days before a dangerous storm, you can warn people.

Two weeks before, you might have time to get supplies and crews where they need to be.

That gap is what makes a weather study published this week interesting. Researchers are using AI to improve forecasts in a stubbornly difficult window: two to six weeks ahead.

The promise is simple. More time to do something about the weather before it arrives.

I am Alex. Welcome to ShortCu8 by Innov8.

Let's dive in.

Weather becomes harder to predict the further ahead you look. Small differences grow, and a precise forecast eventually becomes unreliable.

But there is still a useful question to ask. Is a coming week more likely to be unusually wet, hot or cold?

The study, published in Nature on October 14, tackles that question. Its main tests focused on conditions during the third and fourth weeks ahead, including weekly temperature and rainfall.

So this is about improving the odds of anticipating a wet week. It cannot tell you exactly when rain will hit your street next month.

The researchers added a correction layer to existing forecasting systems.

It compares past predictions with what actually happened, learns recurring mistakes and adjusts future forecasts. Think of someone who keeps being too optimistic about how long a journey takes. Once you know their pattern, you can make a better estimate from the same information.

The team's open-source software works with forecasts from physics-based models, AI models or a mixture of both. It improves the probabilities those systems produce rather than building an entirely new weather model.

There was also a test that mattered beyond looking back at old weather. In the 2025 AI Weather Quest competition, the team's MicroDuet entry submitted predictions before the weather happened. It ranked first across every weather variable and forecast horizon in the competition's inaugural season. Nature

That is encouraging. A forecast is much more convincing when it has to get its answer in before nature reveals hers.

MIT researcher Judah Cohen, a co-author, describes the practical difference: energy officials have told him that two days' notice of an ice storm may be too little, while two weeks could allow adequate preparation.

Farmers could use better information about upcoming rain or cold to plan work and protect crops. Cohen says Kenya's meteorological office has already expressed interest in using the approach. That is interest in implementation, not evidence that it is already protecting harvests there.

The question is how early a warning could help someone prepare. This study does not establish that the system is ready to provide local warnings here.

And better predictions still need people who can act on them. A useful warning has to reach the right person while there is time to respond.

You may see this described as doubling forecasting ability.

The paper reports doubling a particular AI model's modest skill score. That does not mean every forecast is twice as accurate, or that weather prediction is now solved.

In expert reactions gathered by Science Media Centre Spain, researchers caution that large relative gains can start from a weak baseline. Performance also varies by region, and turning research into a dependable operational service remains a challenge.

I find this kind of AI progress easy to care about. The outcome is something ordinary people can understand: a farmer making a better decision, or a community getting more time to prepare.

The test now is whether those better forecasts can become warnings people receive early enough to use.

We read every reply - just reply to this email and let us know how we can improve !

#AD1

80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.

Read the report

#AD2

If you have paid time left on an individual Granola plan, we'll match it with a Wispr Flow subscription that includes Notetaker and dictation, and add bonus time, up to 12 months total. Sign in or create a Wispr account and submit proof of your plan to check eligibility.

Try Notetaker Free