It's a principle thing. All of the breakthroughs in AI and pattern recognition have been via cross-validation. Stupid to waste time waiting for predictions to pan out over decades, while competitors march on. Not my problem that climate scientists are ignorant about validation techniques.
I understand the principle, but I do have some objections. First of all, while LLMs are also impressive, they have not proven capable of predicting the future yet. Secondly, you don't need to wait for the future predictions to keep improving, but it's good to publish them in order to check later.
The issue is that ENSO has a 3 to 7 yr period so that statistical significance in predicting a pattern will take several decades. What absolutely needs to be done 1st is acknowledgement of cross-validation on these 100+ sites. That will guarantee that the model WILL NOT BE FORGOTTEN in the future
a start pukpr.github.io/results/imag...