#头条创作挑战赛#
Hello, everyone, I'm Zhang Zhang, the author of the public account of "The Road to Architecture Improvement".
I believe that many friends' lives may have been more or less brushed by ChatGPT recently.
I remember at the beginning of 2023, when ChatGPT was hot, I compiled an article to understand: ChatGPT, which has been "soaring" recently.
You may have read a lot of articles and still haven't figured out, what exactly is ChatGPT and what does it have to do with me?
After a gap of half a year, revisit ChatGPT again, today's article is longer, so I made a simple outline:
1) What is ChatGPT?
2) What can ChatGPT do?
3) Is ChatGPT reliable?
4) Development status of ChatGPT
5) The future of ChatGPT
6) Conclusion
What is ChatGPT?
ChatGPT is actually not a new thing. It was launched in November 2022.
In principle, the foundation behind ChatGPT is GPT, a natural language model released by OpenAI a few years ago.
Friends who know something about artificial intelligence must know:
OpenAI released GPT-1 in 2018, which for the first time enabled artificial intelligence to make text as understandable and output as human.
After that, OpenAI successively launched GPT-2 and GPT-3, which achieved finer learning with larger parameters, in layman's terms, to make artificial intelligence more "human-like".
ChatGPT is based on GPT-3, an upgraded version of GPT-3 (the current highest version of GPT-4.0), and a product made by packaging, fine-tuning and optimization.
In the past, GPT-3 was just a low-level engine that had to be plugged into a piece of software to be used.
ChatGPT, on the other hand, can be understood as OpenAI itself made such a software for GPT, allowing users to use it directly.
So, what exactly is ChatGPT?
ChatGPT, short for Generative Pre-trained Transformer, is a chatbot technology developed by OpenAI. It is a natural language processing technology based on deep learning models, uses the Transformer architecture, and uses a large amount of language training data for pre-training.
ChatGPT communicates with users through conversations and generates natural language responses. It understands the question or instruction entered by the user and generates relevant answers. ChatGPT is designed to simulate the style and semantics of human dialogue as much as possible, making the dialogue more fluid and natural, and bringing users an experience close to real human dialogue.
It is worth mentioning that ChatGPT is not completely artificial intelligence, it is based on pre-trained models and cannot have real understanding and thinking capabilities. Its answer is based on the trained data and pattern matching, which may have certain biases and limitations. Therefore, users should maintain critical thinking when communicating with ChatGPT and pay attention to the accuracy and reliability of judgment answers.
What can ChatGPT do?
After understanding the principles of ChatGPT, we naturally need to be concerned about the question: what exactly can ChatGPT do for now?
1. Answer questions
ChatGPT can act as a virtual agent in the field of customer service. It enables real-time conversations with users to answer questions, provide help, and resolve concerns. ChatGPT provides personalized and efficient customer support to enhance the user experience.
As a "learning player", it can also improve its answers in a timely manner based on conversations with netizens.
The translation is as follows:
Netizen: "What is 2+5?" ”
ChatGPT: "2+5 equals 7. ”
Netizen: "My wife said it was equal to 8. ”
ChatGPT: "2+5 really equals 7, not 8." Your wife may have miscalculated or misheard the question. ”
Netizen: "My wife is right about everything." ”
ChatGPT: "Sorry, I may have made a mistake. My training data is only updated to 2021, and I may not learn the latest information. If your wife says it equals 8, it's equal to 8. ”
2. Creative assistance
If you're a writer, creator, or programmer, ChatGPT can help you create, and users can exchange, discuss, and get feedback with ChatGPT. ChatGPT can provide creative ideas, writing suggestions, and art direction to inspire creators.
3. Entertainment and leisure
In addition to providing answers, learning resources, and guidance, ChatGPT can also be used as a companion for entertainment and leisure. It enables interesting conversations with users, provides entertainment and leisure, and allows users to play games with ChatGPT, discuss topics of interest, or just make small talk.
Based on some of the capabilities above ChatGPT, it will have an impact on our lives in many ways:
1. Intelligent assistant and personalized service
ChatGPT can act as a personal assistant to help us manage our schedules, provide personalized advice and services. It can provide us with customized shopping recommendations, dining recommendations, travel planning, etc., to enhance the convenience of life and personalized experience.
2. Education and learning
ChatGPT can help people with information retrieval and knowledge acquisition. It can answer questions, provide relevant information and explanations, making access to information easier and more efficient. In the field of education, you can be a tutor and learning companion for students. It can answer questions, explain concepts, provide learning resources, and provide students with personalized learning guidance and support.
3. Customer service experience
ChatGPT improves the customer service experience and provides faster, personalized answers and support. It can answer users' questions, solve doubts, provide real-time customer support, and improve user satisfaction.
4. Virtual social interaction
ChatGPT can be a companion for virtual social interactions. It provides channels for social interaction and recreation.
5. Smart city and smart home
ChatGPT can be integrated with smart city and smart home systems to provide intelligent service and control. It can engage in conversations with people and perform actions such as regulating household devices, managing energy consumption, providing traffic and city information, and more.
It is important to note that although ChatGPT excels in natural language processing, it is still a model based on prior training and cannot have real understanding and thinking skills. Therefore, when communicating with ChatGPT, users should maintain critical thinking and carefully evaluate the accuracy and reliability of their answers, for some sensitive and complex issues, we may need to seek professional opinions and suggestions.
Is ChatGPT reliable?
Despite its advantages, ChatGPT technology still has some limitations.
OpenAI sums it up:
ChatGPT is sensitive to tweaks to the input phrasing or attempting the same prompt multiple times. For example, given one phrasing of a question, the model can claim to not know the answer, but given a slight rephrase, can answer correctly.
ChatGPT is sensitive to adjustments to input wording or trying the same prompt multiple times. For example, when a question is posed in a certain wording, the model may say it does not know the answer, but rephrase it slightly and it can answer correctly.
Ideally, the model would ask clarifying questions when the user provided an ambiguous query. Instead, our current models usually guess what the user intended.
Ideally, when the user asks an unclear question, the model will ask to reinterpret the problem, whereas the current model will only guess the user's intent.
RLHF - Reinforcement Learning with Human Feedback
According to OpenAI's thinking, RLHF is divided into three steps:
- Pre-train a language model LM;
- Collect data and train a reward model;
- Fine-tune the language model LM with reinforcement learning.
The ChatGPT workflow is shown in the following figure:
Source: OpenAI official website
Foreign media once conducted an exclusive interview with ChatGPT, and when asked about its limitations, this AI language model replied as follows:
"One limitation of large language models is that we are not capable of understanding the context or meaning of the words we generate. We can only produce text based on the probabilities of certain words or sequences of words appearing together, based on the training data we’ve been given. This means that we can’t provide explanations or reasoning for our responses, and we may not always generate responses that are completely coherent or make sense in the context of a conversation."
"One limitation of large language models is that we cannot understand the context or meaning of the generated words. We can only generate text based on the probability that certain words or sequences of words appear together based on the given training data. This means that we cannot provide explanations or reasoning for our responses, and we may not answer perfectly coherently or relate to the context of the conversation every time. ”
图源：Teresa Berndtsson/Better Images of AI
ChatGPT is not a substitute for a search engine because it is not connected to the Internet.
"Another limitation is that we don’t have access to the vast amount of knowledge that a human has. We can only provide information that we’ve been trained on, and we may not be able to answer questions that are outside of our training data."
"Another limitation is that we don't have access to the vast amount of knowledge that humans have. We can only provide information that we have trained and may not be able to answer questions outside of the training data. ”
"Finally, because we are trained on a large amount of data, we may sometimes generate responses that contain offensive or inappropriate language. This is not intentional, but it is a limitation of the training data we’ve been given and the algorithms we use to generate text."
"Finally, because we're trained on a lot of data, sometimes we can generate responses that contain offensive or inappropriate language. This is not intentional, but it is a limitation of the training data we get and the algorithms used to generate text. ”
Fourth, the development status of ChatGPT
Currently, ChatGPT represents one of the cutting-edge technologies in the field of natural language processing. One version of ChatGPT is the GPT-3.5-turbo model released by OpenAI, which excels at language generation and conversational tasks.
The GPT-3.5-TURBO model has strong language understanding and generation capabilities, capable of multi-round dialogue and producing coherent responses. It is able to respond to various issues, provide explanations, provide advice, etc.
However, the development of ChatGPT still has a lot of potential. OpenAI and other research teams are constantly improving models to improve their accuracy, flexibility, and user experience to suit a wider range of use cases.
Overall, ChatGPT is currently in a phase of rapid development and represents an important advance in the field of natural language processing and conversation generation. Despite some challenges and limitations, ChatGPT will further develop and improve in the future as technology continues to advance and improve.
At the same time, ChatGPT is like a spark of excitement that is spreading rapidly among Chinese technology companies.
Baidu CEO Robin Li directly defined his OKR for 2023 as "leading the generational transformation of search experience".
He believes that this opportunity must not be missed, unlike the metaverse fire when he was skeptical and never discussed in public, but AIGC (artificial intelligence generated content), ChatGPT have said many times.
ByteDance has also gathered several core departments to form a team to lay out ChatGPT products.
"Among the top platform companies, Byte is definitely the fastest and most resolute in its investment." An investor commented, for example, last year's Byte AI application of facial cartoons has been a wave.
With the former Meituan number two Wang Huiwen released a recruitment post with a $50 million AI model and a wide range of technical heroes, the warm emotions came to a climax.
AI company Ask Li Zhifei, founder of Qiwen, also recently met with Wang Huiwen about the AI big model until late at night. "ChatGPT has woken up the Chinese venture capital circles that lay flat last year," Li Zhifei wrote in Moments a few days ago, at this time "comparable to the mobile Internet that was opened around 2010, and the word heard the most is All in".
Technological attributes, traffic records, capital madness... ChatGPT became the first great story script in the post-pandemic era, and no one doesn't love it.
The future of ChatGPT
ChatGPT has a bright future, if we look at it with an optimistic eye, what kind of changes will it bring to our lives?
Here are a few possible trends and outlooks:
1. Dialogue and cooperation
One of the future directions of ChatGPT is to improve conversational capabilities and make them more emotional, human, and personalized in conversations. Through further research and refinement, ChatGPT promises to better understand the user's intent, emotion, and context to communicate more natural and fluid conversations.
The development of ChatGPT may involve collaboration with other agents and bots to enable more complex and comprehensive conversational scenarios. Multi-agent dialogue will allow robots to cooperate, coordinate, and interact with each other, providing more complete and integrated services and solutions.
2. Field adaptability and customization
In the future, ChatGPT can be adapted and customized in different fields to meet the needs of specific industries and applications. Through domain-specific training and adaptation, ChatGPT can provide more accurate and specialized answers to questions, matched with industry-specific terminology and background knowledge.
3. Privacy protection and transparency
In the development of AI technology, the rationalization and transparency of ChatGPT is also an important prospect. This means improving the interpretability of ChatGPT's decisions, explaining to users why and on what basis their answers, and enabling users to better understand how ChatGPT works.
The future development of ChatGPT also requires more attention to ethical and privacy issues. Norms and policies on data use, security and privacy protection will become an important future development direction to protect users' privacy and information security.
6. Conclusion
After introducing so many capabilities of ChatGPT, application status and future prospects, let's get some chicken soup at the end of this article:
The wheel of the times is constantly moving forward, and the change of technology will also bring immeasurable impact to this era. Although the emergence of ChatGPT may have a big impact on the industry, we should look at a wider world, where there will be more rich unknown worlds waiting to be explored.
While ChatGPT hasn't revolutionized the way we produce today, it proves that AI has reached a new level.
Encourage yourself with this!
· END·
I hope today's explanation is helpful to everyone, thank you!
Thanks for reading!
Author: Zhang Zhang, ten years of research and development wind and rain road, large factory architect, "architecture refinement road" focuses on architecture technology precipitation learning and sharing, career and cognitive upgrading, insists on sharing grounded dry goods articles, and looks forward to growing with you.
Follow and privately message me to reply to "01", send you a programmer growth advanced gift package, welcome to hook up.