AI进化成人的速度，可能比你想象的还慢

Core Insights - The ultimate goal within the AI community is to achieve AGI (Artificial General Intelligence), which refers to creating AI that is as intelligent as a human [1][4] - A group of leading experts has published a paper providing the first quantitative definition of AGI, indicating that current AI models like GPT-5 score significantly lower than the AGI benchmark [9][11] Group 1: Definition and Measurement of AGI - AGI is defined as an AI that can perform at the level of a well-educated adult [11] - The CHC theory from psychology has been adapted to evaluate AI capabilities, emphasizing that intelligence should be assessed through multiple dimensions [12][13] - The evaluation framework consists of ten core abilities, each contributing 10% to the overall score, including general knowledge, literacy, mathematics, reasoning, working memory, visual processing, auditory processing, reaction speed, long-term memory storage, and long-term memory retrieval [16] Group 2: Performance of Current AI Models - Testing results show that GPT-5 scored 58 out of 100, indicating it is below the AGI threshold, with significant weaknesses in long-term memory storage and retrieval [9][19] - GPT-5 excels in general knowledge, literacy, and mathematics, scoring between 9 and 10 in these areas, while it struggles with long-term memory, scoring only 3-4 [19][21] - The current AI models exhibit a phenomenon termed "ability distortion," where they leverage strengths in certain areas to mask deficiencies in others, creating an illusion of competence [21][28] Group 3: Implications and Future Considerations - The paper serves as a comprehensive diagnostic tool for current AI capabilities, highlighting significant deficiencies in fundamental cognitive abilities [28] - The authors caution that the shortcuts taken by AI developers to cover weaknesses may hinder the path to achieving AGI [28] - The proposed standards for AGI, while potentially flawed, shift the discussion from abstract concepts to concrete issues, prompting the industry to reflect on its goals and shortcomings [30]