The sports anchor is not afraid of unemployment for a while: the study says that the AI model analysis game is "almost guessed."

According to information from IT House on June 6th, a new study by a foreign source, Futurism, today's (6th day) evening, found that mainstream AI models performed poorly in the analysis of professional sports competitions. The objective of the study is to examine the performance of the hot AI model in four areas: perception, reasoning, simulation and autonomous mobility, which it is difficult to accurately assess in existing testing methods. It should be noted, however, that they have not yet been peer-reviewed.

To test the extent to which AI can do this, researchers have made professional sports a new set of benchmarks. The new test, called “Strategic video intelligence”, short SVI-Bench, includes 35,000 hours of basketball, football and hockey scenes, 15 million marked rounds, 15,000 hours of career analysis, 23,000 post-race reports and 10.3 million statistical records.

AI's relatively best performer is the "read the picture" step, which is to judge what the player has done at some point in the game. But even the most basic awareness missions, AI is not reliable.ChatGPTGoogle. Gemini. The average recognition accuracy rate for models such as thousands of questions is about 74 per cent. According to the report, in the context of sports jargon, this level is feared to be beyond the reach of even the compulsory lycée of a young baseball game.

To the causal reasoning, the model's performance is further down. Researchers ask AI to explain why a set of tactics or a round occurs in some way, with an average success rate of only about 40 per cent.

For example, researchers have allowed models to judge the anomalies of the NBA Sun Team player, Cody Martin, a three-pointer. He hit the top of the basket this time, and then fell into the basket, but, Chat,GPT The answer is, "He's got the first three-point shot in this game."

Simulation is just as bad. The researchers asked AI to look for evidence based on a player's track and predict where the player would move. The results showed that even the best-performing models were close to random speculation in judging the players ' next move, and that model performance would continue to deteriorate once the projection extended to a longer route towards the goal or basket.

IT House learned from the report that Lorenzo Toresani, a computer scientist and research co-author at the Northeast University of the United States, stated in a university press release that AI “can't tell you why something happened or what happens next”.

Researchers have also tested the autonomous analytical capabilities of the model, which is tantamount to requiring AI, like a human sports anchor, to make complex judgements in relation to post-race data and trends. The results show that the model has only 5% accuracy on this one.

According to Toresani, a good sports interpreter did more than just describe the picture. They will explain why a set of tactics works, predict what happens next, and determine which moments really matter. But research has shown that AI is quite good at depicting the picture, but it is completely out of control in other capabilities.

Torresani summed up that: “The same capacity gaps can occur in any type of work. The true value lies not in describing what is visible, but in understanding why events develop like this, predict what happens next, judge what is important and suggest what should be done next.”

Advertising statements: The external jump links (including not limited to hyperlinks, 2D codes, passwords, etc.) contained in the text are used to convey more information and save time for selection purposes only for reference purposes, which are included in all IT House articles.