After testing the capabilities of several AI models - OpenAI's o3, Google Gemini Thinking, Anthropic Claude 3.7 Sonnet-Thinking, and DeepSeek-R1- Apple researchers have discovered "fundamental limitations" that could prevent AI from evolving as much as some imagine.
Who's the dunce cap for?££££
In short, while AIs have no trouble answering simple questions, they literally collapse when faced with complex puzzle-type problems. These results are "particularly worrying" according to Apple researchers, who question the evolution of current AI towards the famous AGI (artificial general intelligence), supposedly so intelligent that it could destroy humanity in a few years.
Another piece of information from this research is the waste of energy when AIs desperately search for a solution that they ultimately fail to find. In conclusion, the current approach to AI has likely reached its limits, and tech giants will need to explore new methods if they want to continue the race.