| Huawei lab's AI agent clears all 25 public games of a reasoning benchmark Huawei's Noah's Ark Lab and University College London report in a preprint that their AI agentAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise., Memento 3, cleared all 25 public games of the ARC-AGIAGIArtificial general intelligence: an AI system that can do most economically valuable cognitive work at or above human level. There is no agreed test for it, which is why debates about when it will arrive, and what to do about it, are so contentious.-3 benchmark with a perfect efficiency score of 100. The benchmark drops AI systems into unfamiliar puzzle games whose rules they must work out, and scores how few moves they need compared with humans. Run alone, its base model, Anthropic's Claude Opus 5, averages about 40.7 on that score. The model's weights, the numerical parameters set during training, were never changed. The gain instead comes from the agent writing and correcting its own program that predicts what each game will do next. |