11月12日消息，日前，一篇名为《谷歌是否悄然解决了人工智能领域最古老的两个难题？》（Has Google Quietly Solved Two of AI's Oldest Problems?）的文章在人工智能圈内迅速传播。 作者是加拿大滑铁卢劳里埃大学历史学副教授马克·汉弗莱斯（Mark Humphries），这位研究20世纪北美史的学者近年转向数字人文与人 工智能应用研究。他在Substack平台的专栏《Generative History》中披露：他在谷歌AI Studio中试验的一款神秘模型，展现出"几乎完 美"的手写识别能力，以及"自发的、抽象的、符号化推理"现象。 注：AI Studio界面显示A/B测试 谷歌的AI Studio是一个开放实验平台，用户可在其中测试提示词、比较模型表现。最近一周，一部分用户发现系统会随机生成两份答 案，要求他们选择较优者。这是大型AI实验室在模型上线前常用的A/B测试（用于比较两种或多种方案的效果，从而判断哪一个更优） 方式。由此外界推测，这款正在试验的模型可能是即将发布的Gemini-3。 汉弗莱斯的实验原本只是想验证这款模型在"手写历史文档转录"任务上的表现 ...

Core Insights - The future of AI will be shaped by optimists, with expectations for an AI-empowered human development over the next 5-10 years, focusing on symbolic reasoning, spatial intelligence, emotional intelligence, and AI agent ecosystems [1][2] AI Technology Trends - AI is expected to enhance global society significantly, acting as a productivity amplifier rather than a job replacer, potentially increasing productivity by 25% to 50% [2][3] - Four key trends in AI development are identified: 1. **Symbolic Reasoning**: A method based on logical rules and symbolic representation, essential for AI to think and act [2][3] 2. **Spatial Intelligence**: The ability to understand spatial relationships, enabling AI to learn from physical and biological domains [3] 3. **Emotional Intelligence**: The capacity for AI to recognize and respond to emotions, fostering stronger emotional connections with humans [3] 4. **AI Agents**: The evolution of AI agents that will operate in the background, with minimal direct human interaction [3][4] China's AI Development Potential - The potential for AI to help China become "cool" is emphasized, focusing on three elements: the ability to create excellent products, lead global fashion trends, and develop attractive cities [4][5] - AI is seen as a key driver for enhancing China's global influence, particularly in cultural products and sustainable technology exports [5] - Predictions include significant breakthroughs in hard technology sectors like space exploration and chip manufacturing within five years, positioning China as a leader in AI and sustainable development [5]

Core Viewpoint - The article discusses the introduction of a new method called "Chain of Draft" (CoD) that significantly reduces token usage and inference costs while maintaining accuracy in reasoning tasks, inspired by human problem-solving processes [1][2][4]. Cost Efficiency - CoD reduces token usage by 70-90% compared to the traditional Chain of Thought (CoT) method, leading to lower inference costs. For enterprises processing 1 million reasoning queries monthly, costs can drop from $3,800 (CoT) to $760, saving over $3,000 per month [6][7]. Experimental Validation - Experiments evaluated three types of reasoning tasks: arithmetic reasoning, common sense reasoning, and symbolic reasoning. The accuracy of models like GPT-4o and Claude 3.5 Sonnet improved significantly with CoD, achieving around 91% accuracy in arithmetic reasoning compared to over 95% with CoT [8][9]. - In terms of token usage, CoT generated approximately 200 tokens per response, while CoD only required about 40 tokens, representing an 80% reduction [9]. - CoD also reduced average latency for GPT-4o and Claude 3.5 Sonnet by 76.2% and 48.4%, respectively [10]. Task-Specific Results - In common sense reasoning tasks, CoD maintained high accuracy, with Claude 3.5 Sonnet showing an increase in accuracy under CoD conditions [12]. - For symbolic reasoning tasks, CoD achieved 100% accuracy while significantly reducing both token usage and latency [14]. Limitations - The effectiveness of the CoD method significantly decreases in zero-shot settings, indicating potential limitations in its application [16]. - For smaller models with fewer than 3 billion parameters, while CoD still reduces token usage and improves accuracy, the performance gap compared to CoT is more pronounced [18].