A ritual style of artificial intelligence from Google passed its first main expectations

Although the gym of the gym that is largely amounting to deviant generators attracts most of the attention in the field of artificial intelligence, there are sometimes some actually useful technology applications, such as using Google Deepmind for weather intelligence models to predict hurricanes. The experimental tool, which was launched earlier this year, managed to provide accurate modeling of the Erin hurricane as it started getting Steam in the Atlantic Ocean earlier this month.

As ars technica It was first reported that Hurricane Irene – which reached the Category 5 and caused some damage to the island of Bermuda, parts of the Caribbean Sea, and the eastern coast of the United States – presented the Google DeepMind weather laboratory with the first real test of its capabilities.

According to James FranklinThe former head of the specialized unit in the hurricane at the National Hurricane Center, was working well, as it outperformed the official model of the National Center of the National Center and tops many other physics -based models during the first 72 hours of modeling. It eventually fell a little whenever the prediction effort is long, but it is still I topped the consensus model Through expectations for five days.

Although the Google model was impressive in the early days of modeling, the last models are the most important for experts, For all ARS TechnicaIt is the days of three to five of the form that are the officials of making decisions regarding evacuation calls and other preparatory efforts. However, it seems that there may be some promise of weather modeling that works in Amnesty International-although the size of the sample here is very small.

Most of the current golden modeling techniques used in the predictions of storms use physics -based predictions, which mainly try to re -create atmosphere conditions through collaboration in things such as moisture, air pressure and temperature changes to simulate how the storm can act. Instead, the Google model is withdrawn from a huge amount of data trained on it, including “Reinalysis Data Data set that is rebuilding throughout the entire earth from millions of notes, and a specialized database that contains major information about the path, intensity and the size of the winds that number approximately 5,000 hurricanes allocated from 45 years.”

According to GoogleShe tested her model on storms from 2023 and 2024, and Find To predict it for five days managed to predict the storm path with more accurately than most of the other models, which comes about 140 km or 90 miles closer to the final location of the hurricane than the European Center for the Mediterranean Weather Group model, which is the most accurate model available. Now it can refer to a storm that follows in actual time as evidence of the concept, although there is no reason to think about artificial intelligence tools like this will replace other methods at this stage completely.