允中 发自 凹非寺 量子位 | 公众号 QbitAI 在推荐、广告场景，如何利用好大模型的能力？这是个很有挑战的命题。 背后主要有两个核心难点： 1）LLM虽然具备丰富的世界知识和推理能力，但缺乏电商领域的专业知识，在直接应用中往往表现欠佳。 2）LLM的交互方式多为文本，而直接将用户历史行为以文本格式描述会导致输入信息冗长、信息密度低等问题，对建模和推理都造成了困 扰。 为了解决以上问题，阿里妈妈提出了一种 世界知识大模型URM ，通过知识注入和信息对齐，让LLM成为兼顾世界知识和电商知识的专家。相 比于传统的推荐模型，URM通过对用户兴趣的全面理解，可实现基于推理认知能力的用户兴趣推荐。 为了在低时延、高QPS要求的实际系统中上线应用，阿里妈妈技术团队设计了一套面向用户行为动态捕捉的异步推理链路。 目前，URM已经在阿里妈妈展示广告场景上线，在商家的投放效果和消费者的购物体验等指标上均带来了显著提升。 以下面这个例子为例，一个对嵌入式家电、收纳用品有过历史行为的用户，系统推测用户在关注装修且处于硬装的早期阶段，且根据点击商品 推断用户比较注重生活品质，因此推荐了一些全屋定制类产品以及高品质的家电。 在传统 ...

Core Insights - OpenAI is enhancing its ChatGPT Search tool to improve online shopping experiences by providing personalized product recommendations based on user queries and preferences [1][3] - The new feature allows users to ask specific questions in natural language, generating tailored results that include product images, reviews, and direct purchase links [1][2] - ChatGPT's shopping recommendations are based on user memory and aggregated product reviews from the web, without including advertisements [2][3] Group 1: Features and Functionality - Users can input detailed queries, such as budget and specifications, to receive customized product suggestions [1][2] - The "memory mode" for ChatGPT Pro and Plus users will track historical interactions to refine future recommendations [1][3] - Currently, the shopping feature is limited to specific categories like fashion, beauty, home goods, and electronics, with plans to expand [2][3] Group 2: Competitive Landscape - OpenAI's shopping feature is a strategic move to compete with Google, which has a significant presence in online shopping [3][4] - Other AI search platforms, such as Perplexity, are also introducing shopping functionalities, indicating a growing trend in AI-driven e-commerce [3][4] - OpenAI's CEO has shifted from opposing ads in ChatGPT to considering tasteful advertising as a potential revenue stream [3][4] Group 3: Business Opportunities and Growth - The introduction of shopping features presents new commercial opportunities for OpenAI, potentially enhancing user engagement and revenue [4][5] - ChatGPT is currently the most widely used AI tool, with a monthly traffic of 4.7 billion visits and over 20 million subscribers, generating significant revenue [4][5] - OpenAI's revenue reached $3.7 billion last year, with projections indicating that total revenue could exceed $125 billion by 2029 [5]