Workflow
大语言模型(LLM)
icon
Search documents
我悟了如何与AI说话!谷歌 69 页官方提示词秘籍全解析,中文版免费下载
AI科技大本营· 2025-04-22 10:26
(You don't need to be a data scientist or a machine learning engineer – everyone can write a prompt.) 作者 | 王启隆 出品 | CSDN(ID:CSDNnews) 最近,Google 官方发布了一份长达 69 页的 【Prompt Engineering 白皮书】 ,可以说是目前最系统、最权威的"AI 沟通指南"了。我们也是第一时 间翻译好了这本书,准备 【免费】 送给大家! 怎么拿?很简单, 看完这篇文章,参与文末的小活动就行! 现在咱们聊聊,为啥这份白皮书突然就刷屏了?为啥说它是"必学秘籍"? 你不必是数据科学家或机器学习工程师——人人都可以编写提示词。 你苦口婆心解释半天,它抓着一个无关紧要的词就开始自由发挥…… 你想要个 A,它自信满满地给你个 B,还附赠一套又臭又长、看似完美的错误逻辑…… 同一个问题,昨天它懂你,今天它就装傻,效果全看"缘分"…… Google 这份白皮书,不是某个博主的心得体会,不是零散的技巧合集,而是 Google 官方基于对大语言模型(LLM)的深刻理解,系统性梳理出来的 ...
一堂「强化学习」大师课 | 42章经
42章经· 2025-04-13 12:02
吴翼: RL 是机器学习这个大概念下一类比较特殊的问题。 曲凯: 今天我们请来了国内强化学习 (RL) 领域的专家吴翼,吴翼目前是清华大学交叉信息研究院 助理教授,他曾经在 OpenAI 工作过,算是国内最早研究强化学习的人之一,我们今天就争取一 起把 RL 这个话题给大家聊透。 首先吴翼能不能简单解释一下,到底什么是 RL? 传统机器学习的本质是记住大量标注过正确答案的数据对。 举个例子,如果你想让机器学习能分辨一张图片是猫还是狗,就要先收集 10000 张猫的照片和 10000 张狗的照片,并且给每一张都做好标注,让模型背下来。 上一波人工智能四小龙的浪潮其实都以这套框架为基础,主要应用就是人脸识别、指纹识别、图 像识别等分类问题。 这类问题有两个特点,一是单一步骤,比如只要完成图片分辨就结束了;二是有明确的标准答 案。 但 RL 很不一样。 RL 最早是用来打游戏的,而游戏的特点和分类问题有两大区别。 第一,游戏过程中有非常多的动作和决策。比如我们玩一个打乒乓球的游戏,发球、接球、回 球,每一个动作都是非标的,而且不同的选择会直接影响最终的结果。 第二,赢得一场游戏的方式可能有上万种,并没有唯一的标准答 ...
一堂「强化学习」大师课 | 42章经
42章经· 2025-04-13 12:01AI Processing
曲凯: 今天我们请来了国内强化学习 (RL) 领域的专家吴翼,吴翼目前是清华大学交叉信息研究院助理教 授,他曾经在 OpenAI 工作过,算是国内最早研究强化学习的人之一,我们今天就争取一起把 RL 这个话题 给大家聊透。 举个例子,如果你想让机器学习能分辨一张图片是猫还是狗,就要先收集 10000 张猫的照片和 10000 张狗 的照片,并且给每一张都做好标注,让模型背下来。 首先吴翼能不能简单解释一下,到底什么是 RL? 上一波人工智能四小龙的浪潮其实都以这套框架为基础,主要应用就是人脸识别、指纹识别、图像识别等 分类问题。 吴翼: RL 是机器学习这个大概念下一类比较特殊的问题。 传统机器学习的本质是记住大量标注过正确答案的数据对。 所以我觉得人生有一个很好玩的地方是,你需要花很多时间先探索自己的奖励函数是什么,很多人可能努 力了很长时间,最后却发现找错了奖励函数。 这类问题有两个特点,一是单一步骤,比如只要完成图片分辨就结束了;二是有明确的标准答案。 但 RL 很不一样。 RL 最早是用来打游戏的,而游戏的特点和分类问题有两大区别。 第一,游戏过程中有非常多的动作和决策。比如我们玩一个打乒乓球的游戏, ...
一堂「强化学习」大师课 | 42章经
42章经· 2025-04-13 12:01
曲凯: 今天我们请来了国内强化学习 (RL) 领域的专家吴翼,吴翼目前是清华大学交叉信息研究院助理教授,他曾经在 OpenAI 工作过,算是国内最早研究强化学 习的人之一,我们今天就争取一起把 RL 这个话题给大家聊透。 首先吴翼能不能简单解释一下,到底什么是 RL? 因此,RL 其实更通用一些,它的逻辑和我们在真实生活中解决问题的逻辑非常接近。比如我要去美国出差,只要最后能顺利往返,中间怎么去机场、选什么航 司、具体坐哪个航班都是开放的。 但 RL 很不一样。 RL 最早是用来打游戏的,而游戏的特点和分类问题有两大区别。 第一,游戏过程中有非常多的动作和决策。比如我们玩一个打乒乓球的游戏,发球、接球、回球,每一个动作都是非标的,而且不同的选择会直接影响最终的结 果。 第二,赢得一场游戏的方式可能有上万种,并没有唯一的标准答案。 所以 RL 是一套用于解决多步决策问题的算法框架。它要解决的问题没有标准答案,每一步的具体决策也不受约束,但当完成所有决策后,会有一个反馈机制来评 判它最终做得好还是不好。 吴翼: RL 是机器学习这个大概念下一类比较特殊的问题。 传统机器学习的本质是记住大量标注过正确答案的数据对。 ...
杨立昆“砸场”英伟达:不太认同黄仁勋,目前大模型的推理方式根本是错的,token 不是表示物理世界的正确方式|GTC 2025
AI科技大本营· 2025-03-21 06:35
责编 | 王启隆 出品丨AI 科技大本营(ID:rgznai100) 黄教主的演讲 感觉才没过几天,今年的 GTC 英伟达大会也即将迎来尾声了。 而今年比尔·达利则是对话"AI 教父" 杨立昆 (Yann LeCun),很有前后呼应的感觉。 但 GTC 并不只有黄仁勋和杨立昆,还有许多精彩的演讲与对话,比方说: ………… 接下来的一段时间, CSDN AI 科技大本营 将会在「 GTC 2025 大师谈 」栏目持续更新这些精华内容的全文整理,尽情期待。 比尔·达利 自己就在采访杨立昆之后进行了一场 演讲 ,系统性地讲解了英伟达 2024 一整年的四大项目进展,内容干货很多; OpenAI o1 作者 诺姆·布朗 (Noam Brown)和英伟达的 AI 科学家来了一场 对话 ,他认为现在 AI 圈最需要来一场革命的,就是这些五花八 门的 基准测试 (Benchmark),而且改这个东西还不需要花太多算力资源; 2018 年诺贝尔化学奖得主 弗朗西斯·阿诺德 (Frances Arnold)围绕 AI for Sciense 还有蛋白质工程进行了一场相当硬核的 圆桌对话 ; UC 伯克利教授 彼得·阿比尔 (P ...
GenAI 的存储解决方案 第 7 部分:解决方案梳理
Counterpoint Research· 2025-03-18 09:14
Group 1 - The importance of cost factors such as power consumption, footprint, and price in addition to advantages like bandwidth and capacity for GenAI solutions [1] - The configuration of memory, being a passive component reliant on the CPU, will change according to the processor [1] - ARM's recent strategic adjustments and potential changes in MoE-based large language models (LLMs) may lead to shifts in future architectures and solutions [1]
GenAI 的存储解决方案 第 7 部分:解决方案梳理
Counterpoint Research· 2025-03-18 09:14
对于 GenAI 的解决方案而言,诸如带宽和容量之类的优势固然重要,但功耗、占用面积和价格等成 本因素也需要加以考量。由于存储器是一种依赖于中央处理器(CPU)的被动组件,其配置会根据 处理器而做出改变。在这方面, ARM 近期的战略调整以及基于 MoE 的大语言模型(LLM)可能出 现的变化,都可能会导致未来架构和解决方案发生改变。 Haylee Xu / 市场专员 电话: +86 15959754429 邮箱:haylee.xu@counterpointresearch.com 数据来源:Source: Counterpoint Research 点击阅读原文下载完整版 PDF 报告 Rick Cui / 客户服务总监 电话: +86 13801127537 邮箱:rick@counterpointresearch.com 媒体采访 业务咨询 ...
彭博数据洞察 | 透过AI看新闻,投资信号抓得准
彭博Bloomberg· 2025-03-14 03:08
Group 1 - The article emphasizes the importance of AI-driven news summarization to extract insights and signals from real-time news, which has become a critical intelligence source for quantitative investors [3][4] - Bloomberg's flagship product provides comprehensive support for news headlines and content, covering thousands of themes and regions, with a rich tagging system to label topics, securities, and individuals [3][4] - The article illustrates the impact of news events on market prices, using the example of the Keystone pipeline shutdown, which led to a significant increase in crude oil prices shortly after the news broke [3][4] Group 2 - The article discusses the release of a framework by the Taskforce on Nature-related Financial Disclosures (TNFD) aimed at helping companies and financial institutions assess and disclose their reliance on natural resources and environmental impacts [7][8] - It highlights the importance of understanding ecological interconnections for investors and companies, as these factors can significantly affect market performance, brand reputation, and compliance status [7][8] - The example of Meiji Holdings illustrates how integrating supply chain data with biodiversity databases can help identify risks associated with suppliers located in high water stress or biodiversity integrity areas [8][9] Group 3 - The article analyzes the European automotive industry, indicating that sales momentum has been declining, with signs of demand weakness among suppliers emerging before the broader market recognized the trend [11][12] - The analysis is based on Bloomberg's global supply chain database, covering over 1,500 suppliers in the European automotive sector across 53 countries, combined with timely standardized financial data [12] - This integration of financial data and supply chain information is crucial for predicting industry trends and optimizing decision-making [12]
DeepSeek对英伟达长期股价的潜在影响
CHIEF SECURITIES· 2025-03-12 06:38
Investment Rating - The report does not explicitly provide an investment rating for the industry or specific companies involved. Core Insights - DeepSeek's significant cost advantages in training and inference have led to substantial market impacts, including a notable drop in Nvidia's stock price and market capitalization [2][11][12] - The introduction of DeepSeek's models has the potential to disrupt existing AI companies by lowering the barriers to entry for smaller firms and individuals, thereby increasing overall demand for computational resources [15][16] Summary by Sections Section on DeepSeek's Market Impact - DeepSeek achieved the top position in download rankings on both the Chinese and US App Store, coinciding with a major drop in the semiconductor index and Nvidia's stock [2] - Nvidia's market value decreased by nearly $600 billion, marking one of the largest single-day market cap losses in history [2] Section on Cost Structure - DeepSeek's training costs for their V3 model were reported to be under $6 million, utilizing approximately 2000 H800 GPUs [6][7] - The inference cost for DeepSeek's models is significantly lower than that of OpenAI, with DeepSeek charging only 3% of OpenAI's rates for similar token inputs and outputs [7][9] Section on Training Innovations - DeepSeek implemented innovative training strategies that reduced costs, particularly by optimizing the supervised fine-tuning (SFT) process [9][10] - The team utilized pure reinforcement learning (RL) without human feedback, achieving performance comparable to OpenAI's models [9][10] Section on Future Implications for AI Industry - DeepSeek's advancements may lead to increased competition among AI firms, particularly those relying on self-developed large models [12][13] - The report suggests that while Nvidia's stock may have been negatively impacted in the short term, the overall demand for their chips could increase as AI commercialization accelerates [14][16]
2025中国AI“奇点”已至?摩根大通:应用井喷在即,DeepSeek点燃算力需求,阿里或成最大赢家
硬AI· 2025-03-10 10:32
Core Viewpoint - Morgan Stanley believes that China's Generative AI (GAI) development is at the beginning of its second phase, with Alibaba positioned as a key player in the Infrastructure as a Service (IAAS) value chain, likely to outperform peers in this phase and potentially benefit from the third phase of applications [2][3]. Phase Summaries - **Phase 1**: Development of large language models (LLMs), focusing on building and optimizing LLMs [5]. - **Phase 2**: Application of GAI in existing applications and services, currently at the beginning stage where companies are exploring value creation models [6]. - **Phase 3**: Surge in internet service consumption as GAI applications become widespread, leading to significant financial gains for internet operators [7]. - **Phase 4**: Emergence of native GAI "killer applications" that will fundamentally change market competition and introduce new business models [8]. Infrastructure as a Service (IAAS) Insights - The report highlights that companies in the IAAS value chain will perform well in the second phase of GAI development, with revenue expectations likely to be positively revised. Alibaba is identified as the most promising stock in this area [11][12]. - Tencent, while holding a significant market share in China's IAAS, is viewed more as a beneficiary of AI applications rather than a primary player in the IAAS space [12][13]. Other Key Players - **Kuaishou**: Identified as an undervalued AI beneficiary, with expectations that AI will significantly enhance user engagement and monetization capabilities. The AI video/image generator Kling is projected to have over 6 million users by December 2024, indicating substantial monetization potential in various commercial verticals [15]. - **Baidu**: Positioned as both an IAAS cloud value chain player and a potential GAI application beneficiary. The company's stock outlook is contingent on the transformation of its core advertising and cloud business narratives [16].