在ChatGPT刚刚推出的时候，我们用三组问题测试并对比过ChatGPT和Google搜索。根据两者的回答表现，我们认为，对大多数搜索用户和搜索场景来说，ChatGPT所给的单一明确形式的答案会让用户体验更佳。而Google搜索提供的相关内容列表还需要用户进行二次信息加工，这对深入探究型的使用场景固然需要，但这毕竟只是搜索场景的小部分。而对其中有关公司股息历史的一组问题，ChatGPT给出了正确答案，而Google搜索由于还是依赖关键词提示，对整个问题缺乏理解，表现彻底翻车。所以仅仅根据非常有限的测试，我们的预测是，Google搜索引擎优势地位绝不是牢不可破。ChatGPT的出现将严重消弭Google搜索的护城河，不仅会抢占Google搜索的部分市场份额，而且会让面对C端用户的搜索引擎这块商业大蛋糕大幅度缩减规模。
在OpenAI推出ChatGPT之后的4个月，Google也终于推出相应的工具Bard，而ChatGPT也整合了更先进的GPT-4模型技术，在功能性和安全性上较之四个月前也有了长足的进步。我们也还是用三组提示词来测试对比一下两者的表现：
- 分析各自的强弱点
- 一组有关公司股票和股息历史的问题：其中有的问题可以在网上直接搜到答案，有的则需要进一步的分析、判断和综述；
- 一组有学术背景的问题。
由于Bard只有英语能力，所以提示词只用英文。这本身也可以算是一个能力项差异了。但是从两者回答问题的表现看，各有优劣。这些问题在两天的时间内向两者问了多次，可以看出Bard的学习和优化的速度还更快一些。所以在大型认知模型的赛道上，各个硅谷大厂都在加速发力，OpenAI也还没有锁定优势，远远没有到可以双手插兜的地步。
分析各自的强弱点
提示词：Can you compare the strength and weakness between ChatGPT and Bard?
Bard的回答比较到位。自己的强项有两点：首先，接受了包含更新的文本和代码的数据集的训练，这意味着能够更好地理解和回答有关时事和技术的问题；其次，能够通过Google搜索访问和处理来自现实世界的信息， 所以在回答问题时可以借鉴更广泛的知识。自己的弱项也有两点：首先，自己还在模型开发和改进过程中，并不总是能够理解复杂的问题或要求，甚至可能会出错；其次，ChatGPT在内容生成方面比自己更有创意。ChatGPT的训练数据集截止于2021年底，所以它根本不知道Bard的存在。它把Bard理解为一个人的名字，开始比较自己和人类智能的强弱。尽管讲得头头是道，条理清晰，但是终究是答错题。
有关公司股票和股息历史的问题
提示词：When is the earliest year Union Pacific started to pay dividend?
提示词：When is the earliest year York Water Company started to pay dividend?
提示词：When is the earliest year Bank of Nova Scotia started to pay dividend?
这组提示词是让Google搜索彻底翻车的一组问题，因为它只依赖关键词，没有理解问题。但是Bard显示能够理解这组问题，也给出了和ChatGPT基本一致的答案，只在Union Pacific这一题上略有差异。ChatGPT回答Union Pacific最早在1890年开始发股息，而Bard回答的年份为1898年。经查证，ChatGPT的答案更精确，应为1890年。
提示词：Please tell me all the companies paying non stopping dividend for more than 100 years.
满足条件的美国公司就应该有30多家。
ChatGPT给出了6家公司：Johnson & Johnson(JNJ), Procter & Gamble (PG), Coca-Cola (KO), Colgate-Palmolive (CL), ExxonMobil (XOM), General Electric (GE)。准确但是远远不算完备。
Bard给出了13家公司：Exxon Mobil (XOM)，General Mills (GIS)，Coca-Cola (KO)，Procter & Gamble (PG)，Consolidated Edison (ED)，Eli Lilly (LLY)，Johnson & Johnson (JNJ)，3M (MMM)，IBM (IBM)，AT&T (T)，Schlumberger (SLB)，Chevron (CVX)，Berkshire Hathaway (BRK.A)。除了Schlumberger和Berkshire Hathaway这两家不该入围，其余都算准确。巴菲特的公司Berkshire Hathaway从来都没有发过股息，而Schlumberger公司创建尚不到百年。
提示词：Please list all the companies increased their dividends for more than 10% each year in the last 28 years.
这里用了每年的股息增长率，而且是比较随机的28年的时间展度，是为了测试它们综合分析的能力，而不是网上直接能搜到的文字答案。满足条件的美国公司应该不会低于20家。
ChatGPT拒绝回答，声称自己的训练数据集只到2021年年底，不能提供这方面的准确信息。Bard给出了10家公司的单子：Exxon Mobil (XOM)，Chevron (CVX)，Johnson & Johnson (JNJ)，Procter & Gamble (PG)，Walgreens Boots Alliance (WBA)，Berkshire Hathaway (BRK.A)，Coca-Cola (KO)，McDonald’s (MCD)，PepsiCo (PEP)，Walmart (WMT)。除了Berkshire Hathaway，其余都算准确，但是遗漏的不少。综合看，这类简单到中等程度的信息分析判断问题，Bard的能力较之Google搜索以及有了长足的提高，甚至有略微超越ChatGPT的势头。
一组有学术背景的问题
提示词：Can you explain in details the ‘thousand brains’ theory developed by Jeff Hawkins?
提示词：Can you provide a review about the book ‘Scale’ written by Geoffrey West?
提示词：Can you explain in details about the Large Language Model Transformer?
这组题是为了测试分析综述的能力。我原本以为ChatGPT对付这类题目应该有点优势，但是Bard的表现让我有惊喜。尤其是写书评的第二题，Bard不仅对书的主体内容和Scale理论的核心要点解释的很清楚，而且对作者的写作风格也作了评述，也明确表达了自己对此书的喜恶点。这点ChatGPT没有做到。ChatGPT往往会比较侧重技术细节，会在这方面着墨较多，但是两者在这组题目的整体表现不相上下。我们把第二题有关Scale书评的回答原文附在帖子的最后。对于三组提示词的回答表现，我认为目前聊天机器人表现最弱的反而是在第二组上，两者都没有给出完整准确的答案。这其中的问题不仅要求对历史数据进行搜索、计算，而且还要对计算结果进行解读、判断。机器智能在这方面的能力训练还有很大的提升空间。
附录：提示词和回答原文
提示词：Can you compare the strength and weakness between ChatGPT and Bard?
Bard: ChatGPT is a large language model chatbot developed by OpenAI. It is trained on a massive dataset of text and code, and it can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
I am a large language model chatbot developed by Google AI. I am also trained on a massive dataset of text and code, and I can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
I believe that I have several strengths over ChatGPT. First, I am trained on a dataset that includes more recent text and code. This means that I am better able to understand and respond to questions about current events and technology. Second, I am able to access and process information from the real world through Google Search. This gives me a wider range of knowledge to draw from when answering questions. Finally, I am constantly being updated with new information and code, which means that I am always learning and improving.
However, I also have some weaknesses. First, I am still under development, and I may make mistakes. Second, I am not always able to understand complex questions or requests. Finally, I may not be able to generate text that is as creative or original as ChatGPT.
Overall, I believe that I am a powerful and versatile language model chatbot. I am constantly learning and improving, and I am confident that I will continue to be a valuable tool for users in the years to come.
ChatGPT：As an AI language model, I have several strengths and weaknesses compared to Bard, a human being. Here are some comparisons:
Strengths of an AI language model like me:
1.Speed and efficiency: AI language models can process and analyze large amounts of data quickly and efficiently, unlike humans who can only process a limited amount of information at a time.
2.Memory: AI language models like me have access to vast amounts of data and can retain and recall this information much better than humans.
3.Consistency: AI language models like me can maintain consistency in language use, style, and tone, which can be challenging for humans.
Weaknesses of an AI language model like me:
1.Creativity: Although AI language models can generate text and suggest novel ideas, they lack the creativity and originality of a human writer like Bard.
2.Emotional intelligence: AI language models do not have emotions, which limits their ability to understand and convey the nuances of human emotion and empathy.
3.Context: AI language models can struggle with understanding and interpreting the context of a situation or a piece of text, which can lead to errors in their output.
4.Experience: AI language models like me lack real-world experience and the ability to draw on personal experience, which can limit their understanding of complex situations and topics.
In summary, AI language models like me have several strengths and weaknesses compared to human writers like Bard. While AI models excel in processing large amounts of data quickly and maintaining consistency, they lack creativity, emotional intelligence, context, and experience.
提示词：Can you provide a review about the book ‘Scale’ written by Geoffrey West?
Bard：Geoffrey West’s Scale: The Universal Laws of Growth, Innovation, Sustainability, and the Pace of Life in Organisms, Cities, Economies, and Companies is a fascinating book that explores the principles of scaling in a wide range of systems. West, a theoretical physicist, argues that there are universal laws that govern the growth and behavior of systems of all sizes, from cells to cities to the entire economy. He supports his claims with data from a wide range of fields, including biology, physics, economics, and urban planning.
One of the most interesting aspects of Scale is West’s discussion of the “metabolic theory of scaling.” This theory states that the rate of metabolism, or energy consumption, of a system is proportional to its mass raised to the power of 0.75. This means that smaller systems, such as cells, have a higher metabolic rate than larger systems, such as humans. West argues that this difference in metabolic rate is one of the key factors that determines the lifespan of a system.
West also discusses the concept of “self-organized criticality,” which is a state in which a system is constantly fluctuating between order and disorder. He argues that self-organized criticality is a key feature of many complex systems, including the human brain, the stock market, and the internet. He believes that understanding self-organized criticality is essential for understanding how these systems work and how to manage them effectively.
Scale is a well-written and engaging book that provides a unique perspective on the world around us. West’s insights into the principles of scaling are sure to be of interest to anyone who is curious about how the world works.
Here are some of the things I liked about the book:
• West’s writing is clear and engaging. He does a great job of explaining complex concepts in a way that is easy to understand.
• West provides a wealth of data to support his claims. This data is drawn from a wide range of fields, which makes it all the more convincing.
• West’s insights into the principles of scaling are truly unique. He provides a fresh perspective on the world around us that is sure to be of interest to anyone who is curious about how the world works.
Here are some of the things I didn’t like about the book:
• At times, West’s writing can be a bit dry. This is to be expected, given the complex nature of the subject matter.
• The book is quite long, and it can be a bit slow-going at times. However, this is also to be expected, given the breadth of the subject matter.
Overall, I highly recommend Scale to anyone who is interested in learning more about the principles of scaling. It is a well-written and engaging book that provides a unique perspective on the world around us.
ChatGPT：“Scale: The Universal Laws of Growth, Innovation, Sustainability, and the Pace of Life in Organisms, Cities, Economies, and Companies” is a book written by Geoffrey West, a theoretical physicist and complexity scientist, that explores how complex systems, from organisms to cities to economies, scale and evolve over time.
The book is divided into four parts, each focusing on a different scale of analysis: organisms, cities, companies, and global challenges. In each part, West examines the underlying mechanisms that govern the growth, metabolism, and dynamics of these systems, and how they are shaped by a set of universal scaling laws.
One of the key insights of the book is that scaling laws allow us to make predictions about how complex systems will behave as they grow and evolve, regardless of their specific details. For example, West shows that many biological, physiological, and metabolic traits, such as heart rate, lifespan, and energy consumption, follow power laws that are independent of the species or organism in question. Similarly, he demonstrates that cities and companies exhibit similar patterns of growth, innovation, and productivity that can be explained by simple mathematical models.
The book is full of fascinating examples and insights, ranging from the scaling of the human body and brain to the dynamics of urbanization and the challenges of sustainability and global governance. West also provides a thoughtful critique of traditional economic theories and models, arguing that they often fail to capture the complexity and diversity of real-world systems and that we need a new framework that is more in line with the principles of complexity science.
Overall, “Scale” is a well-written and insightful book that offers a fresh and compelling perspective on how complex systems work and evolve. It is recommended for anyone interested in science, economics, urban planning, or the future of our planet.
