.

Great use of AI. Hope to see some disease cures next.

For a long time, Weather.com touts itself as run by Watson.

I would have assumed that all models are AI in some sense of the word.

So...

First of all, if we go way back (29 Apr 2026), old research showed that physics models outperformed AI models

Physics-based models outperform AI weather forecasts of record-breaking extremes
https://www.science.org/doi/10.1126/sciadv.aec1433

QuoteHere, we show that for record-breaking weather extremes, the physics-based numerical model High RESolution forecast (HRES) from the European Centre for Medium-Range Weather Forecasts still consistently outperforms state-of-the-art AI models GraphCast, GraphCast operational, Pangu-Weather, Pangu-Weather operational, and Fuxi. We demonstrate that forecast errors in AI models are consistently larger for record-breaking heat, cold, and wind than in HRES across nearly all lead times.

The question remains whether or not this is a pattern, or a lucky guess.

So, what is the difference?

https://rss.org.uk/news-publication/news-publications/2026/general-news/cctf-publish-new-explainer-artificial-intelligence/

QuoteTraditional weather forecasts rely on physics-based models that simulate the atmosphere. These models solve huge sets of equations about how air, moisture and energy move around the globe. They are powerful but computationally demanding, requiring the world's fastest supercomputers. Artificial intelligence models, in contrast, learn directly from data without using the physics equations.

QuoteAre AI models 'better' than standard weather forecasting models? 

It depends on what you are trying to predict.

[AI models] are especially strong at predicting big, wide patterns, like large storms moving across oceans.


However, physics-based models remain more reliable for predicting local details... When AI models are trained, they are generally rewarded for being close to the observed outcome on average, even if they miss small details. The nature of this rewarding approach can create a bias toward forecasts that look smoother, with fewer small-scale features.

QuoteHowever, once trained, running AI forecasts is far less energy intensive per forecast than traditional high-resolution weather forecasts. Recent estimates suggest that AIFS can reduce energy use by around 1,000 times relative to conventional methods.

QuoteCan AI forecasts be trusted for extreme weather events? 

The evidence on this is mixed but improving. On the positive side, Google's GenCast shows good accuracy for some extreme weather up to around 15 days, and the ECMWF report shows promise for their AIFS model in tracking tropical cyclones.

Gem notes that AI forecasts, since they have no physics model, still sometimes forecast impossible conditions, like a snowstorm in warm weather.

Quote from: Travoli on October 13, 2026, 06:45:56 PMGreat use of AI. Hope to see some disease cures next.

I started keeping a list of "AI for serious shit" in response to a physician friend who kept insisting it was a "toy". Google Docs says the last update was back in March. Here are a couple of highlights from medicine.

These diseases were thought to be incurable. Now AI is unlocking new treatments
https://www.bbc.com/future/article/20260309-ai-is-finding-treatments-for-incurable-diseases

"Discovery of a structural class of antibiotics with explainable deep learning," Felix Wong et al., Nature (2023)
https://www.nature.com/articles/s41586-023-06887-8

Johns Hopkins researchers develop life-saving AI for early sepsis detection
https://www.news-medical.net/news/20260512/Johns-Hopkins-researchers-develop-life-saving-AI-for-early-sepsis-detection.aspx

That is for a system developed in 2022. For perspective, there are 250,000 hospital sepsis deaths annually in the US. Reducing that by 18% (2022 result) would potentially save 45,000 lives per year (I assume the actual number is lower because some of those people will die during that hospital visit due to other issues).

I think the "cure disease" framework is generally the wrong way to look at things, BTW. I think it was a reasonable paradigm when people were finding a vaccine for polio or vitamin that could solve rickets. These were simple, single-cause diseases (not that the solution was "easy" but it was "simple). But it became a detrimental paradigm when applied to "a cure for cancer" or other complex diseases that are not actually A disease.

The sepsis model is interesting. Yes, it only prevented 18% of deaths. But if you find five such interventions, you get rid of 63% of deaths (roughly). And that's mostly what we've seen with cancer.

I think the future of medicine will look more like *treating* cancer than like *curing* rickets.  In other words, I expect that AI will push the frontiers forward by incremental improvements that maybe aren't even observable from one year to the next, but which result in profound improvements one decade to the next. Which is maybe disappointing for most of us, but I suspect our grandchildren will have a completely different medical context (and in my utopian vision of where things are going, a much better one; I also have a dystopian vision of where things are going, but let's not go there)

I don't expect to ever see the AI equivalent of this

>I also have a dystopian vision of where things are going, but let's not go there

Well, I was going to comment that I hate that my tech news feeds feel like doomscrolling now but I guess I won't. ^hhh

> my tech news feeds feel like doomscrolling

I always feared I would live in interesting times.

>AI is unlocking new treatments

This is all fantastic. It's always nice to see spots of light peaking through the doom.