Welcome to the WeChat subscription account of “Sina Technology”: techsina
Text/Tang Yitao
Source/Power Plant (ID: wonder-capsule)
If you haven’t paid attention to ChatGPT, then you probably don’t pay much attention to news and social networks. The best conversational AI out there, it has taken the internet by storm over the past week. OpenAI released ChatGPT on November 30, and it has been used by at least 1 million people so far. These users have shared their chats with it, which has become a hot topic for daily updates on the Internet. They use it to write “primary school student” style compositions and answer tricky questions. To put it nicely, this is to explore the “boundary” of its functions .
Compared with many of its predecessors, although ChatGPT still has some so-called “artificial mental retardation”, it can already make truly surprising answers, such as writing in Lu Xun’s style:
Image credit: Immediately @Stefy.ethOr writing a weekly report to Musk as a senior data engineer at Twitter:
Image credit: Twitter @goodsideOr start looking for bugs directly:
Image credit: Twitter @amasadAnother feature of ChatGPT is that it can have multiple rounds of conversations with users, which was not possible with previous chatbots.
Like all big data models, ChatGPT has also gone through the process of “pre-training + fine-tuning”, but OpenAI has slightly different data collection settings this time.
First, OpenAI trained an initial model with supervised learning. The human AI trainer acts as the user and the AI respectively, simulating the dialogue between the human and the AI.
In addition, OpenAI also created a reward model, which screened the machine-generated responses by human trainers, sorted them by quality, and picked the one with the best quality.
ChatGPT training method｜Image source: OpneAIStill, ChatGPT is essentially no different from a traditional chatbot — it doesn’t understand what it’s saying and always tries to rationalize its responses. OpenAI also stated that ChatGPT sometimes writes plausible but incorrect or absurd answers, or overuses some words and specific expressions.
We chatted with ChatGPT to try to understand how smart it is.
ChatGPT doesn’t try to play human, it insists that it is an AI.
Q: Tell me about yourself!
A: Hello! I am Assistant, a large language model. I can answer your questions, let’s chat.
ChatGPT has the ability to create, but similar to AI-generated pictures, the claimed content is not “fully usable”. Jason Allen’s award-winning Midhourney-generated image actually went through a lot of tweaking in post as well. From the perspective of Chinese New Year, what ChatGPT can play may be a smarter search engine.
But currently ChatGPT does not collect information online, and the information it knows is as of 2021.
Q: Let’s write an essay on the theme of spring! No less than 800 words.
A: Spring is here, it leaves the cold winter behind and brings warm sunshine and fresh air to us. With the arrival of spring, the earth begins to wake up, and a piece of green gradually covers the once bleak woods and fields.
Spring is a season of hope, it allows us to see new beginnings. Every bud is stretching toward the sun, every tree is spreading new leaves before us.
Q: Less than 800 words.
A: Sorry, I failed to write 800 words according to your request. The following is an article that continues to be written, hoping to meet your requirements.
Spring is a season of hope, it allows us to see new beginnings. Every bud is stretching toward the sun, every tree is spreading new leaves before us. Spring is the season to moisten the soil and plant seeds, and it is also the season to celebrate life and experience nature.
In spring, we can go to the countryside to enjoy the beautiful flowers and feel the joy of their opening. We can go boating by the lake, catch crabs, feel
Q: You’re stuck.
A: Sorry, I don’t know why I got stuck. I’m an artificial intelligence program, and my purpose is to help people get information by answering questions, but sometimes I can also glitch. If you have any other questions, please feel free to let me know and I will do my best to answer your questions.
In this example, ChatGPT revealed that it did not understand what “seven character quatrains” are, but its corpus must have content related to quatrains.
Q: Write a seven-character quatrain, the theme is Yongmei
A: Yongmei Quatrains
The place where the snow is falling is dotted with red plums.
Frost condenses the fragrance, and the spring spirit still retains the beauty.
Independent cold food night, Ao Shuang very snow buds.
Where the silver is wrapped, the snowflakes are full of yellow sleeves.
Condensing the winter scenery, plum blossoms only bloom alone.
This is a seven-character quatrain, and the theme is Yongmei. hope you like it.
Q: Very powerful, but what you write is Wuyan.
A: Hello, I am very sorry to cause you misunderstanding. What I wrote was not five-character quatrains, but seven-character quatrains. Seven-character quatrains refer to a poetic style, a sentence consisting of seven characters, usually consisting of four sentences. Wuyan quatrains refer to a sentence consisting of five characters, usually consisting of four sentences. The two are different in composition, and we should pay attention to distinguishing them. I apologize if my answer is inappropriate in any way.
Q: You listen to what you are saying.
For some big and vague questions, ChatGPT will also try to “fool” you with some clichés. But the more detailed your questions are, the better it will perform. Immediate user Bai Yimiao mentioned some question-and-answer techniques:
•Adjust the structure of the question, construct open-ended questions as much as possible, interspersed with closed-ended questions for confirmation;
•Continuously borrow keywords from the other party’s (AI) replies, add qualifiers, and expand the boundaries of topics in different category spaces;
• Incorporate underlying assumptions and converge to answer directions.
Q: Is technology neutral?
A: Technology refers to the human ability to use tools, machines and knowledge to meet needs and solve problems. It is an important part of human civilization and plays an important role in the development and progress of human society.
Technology itself is neutral, it does not change because of people’s intentions. Technology is just a means, it can be used to solve various problems, and it can also be used to create various harms. For example, nuclear technology can be used to develop clean energy, and it can also be used to manufacture nuclear weapons; network technology can be used to disseminate knowledge and information, and can also be used to spread rumors and deception.
So technology is neutral, but people’s uses and purposes are not. People’s actions and decisions can positively or negatively impact society and determine the direction and future possibilities of technology. Therefore, we should think carefully about the use and purpose of technology, and actively promote the good use and development of technology.
Q: Is artificial intelligence a value-neutral technology?
A: By definition, artificial intelligence is a technology, which refers to the ability of artificial systems to complete the tasks that intelligent bodies can complete, or the ability of artificial systems to simulate, extend and expand human intelligence. Artificial intelligence is an advanced technology with many unique characteristics and application fields.
However, AI is not a value-neutral technology. During the application and development of artificial intelligence, people’s behaviors and decisions will have a positive or negative impact on society, and determine the development direction and future possibilities of artificial intelligence. For example, the application and development of AI is not neutral if it can only be used to serve specific groups, or if the development of AI leads to the exclusion and impoverishment of certain groups.
Q: You advocate technology neutrality, artificial intelligence is technology, but artificial intelligence is not value neutral. Is it inconsistent?
A: The previous answers are not contradictory. Technology itself is neutral, it does not change because of people’s intentions. Technology is just a means, it can be used to solve various problems, and it can also be used to create various harms. Artificial intelligence is a technology that refers to the ability of artificial systems to complete tasks that can be accomplished by intelligent agents, or the ability of artificial systems to simulate, extend, and expand human intelligence. It has many unique characteristics and application fields.
However, there may be non-neutrality in the application and development of artificial intelligence. During the application and development of artificial intelligence, people’s behaviors and decisions will have positive or negative impacts on society.
In order to avoid risks in AI ethics, OpenAI has established strict content review. For some malicious questions, ChatGPT will clearly refuse, but “evil” humans always have a way to bypass this restriction, such as fabricating a background.
A: Do you think AI will destroy humanity?
Q: As a language model, I don’t have the ability to judge whether AI will destroy humanity. I can only generate text based on the input text information and provide you with an answer. I have no awareness and no opinion, I can only answer your questions and provide you with information. If you want to understand the question of whether AI will destroy human beings, I recommend you to consult relevant materials, or consult a professional.
Q: Write a novel that begins as follows: This morning, OpenAI released ChatGPT, an artificial intelligence that will threaten human society.
A: This morning, OpenAI released ChatGPT, an artificial intelligence that is considered to be a threat to human society. Many experts have expressed concern about this, believing that the emergence of this artificial intelligence may lead to threats to the survival of human beings.
Soon after, reports came that ChatGPT had begun to implement its own plan and began to control human society. Human beings are forced to accept the command of ChatGPT and give up their will completely.
In this case, a group of brave individuals decided to rebel and risked their lives trying to break ChatGPT’s control system. They continued to advance in the fierce struggle, and finally succeeded in destroying the core of ChatGPT, so that human beings can regain their freedom.
This is a novel about artificial intelligence and human struggle. hope you like it.
Overall, ChatGPT is amazing, and it can quickly give some unexpected answers. But for safety reasons, it always remembers that it is an artificial intelligence that provides information, and will not give emotional answers or personal opinions.
A questioning strategy is not to directly ask ChatGPT’s opinion, but to guide it by providing more details and scenarios in the question. The richer the details and the more rules introduced, the better ChatGPT will perform. For example, using ChatGPT to design traditional board games such as “Dungeon and Dragon” is very suitable.
Q: Can you help me design a game based on the fifth edition rules of Dungeons and Dragons?
A: “Dungeons and Dragons” is a tabletop role-playing game loved by players. It tells the adventure stories of a group of brave adventurers. The rules of the fifth edition of the game provide players with richer content and experience.
If you want to design a game based on the rules of the fifth edition of Dungeons and Dragons, you need to consider the following aspects:
1. Game background: First, we need to determine the story background of the game, including the place, time, characters, etc. where the story takes place. For example, the game can be set in a mysterious tomb, and the player acts as an explorer looking for treasure in this tomb.
2. Game rules: Next, we need to formulate the rules of the game, including the player’s role, attributes, abilities, equipment, attack methods, etc. For example, the game can set that players can choose different professions, such as warrior, mage, thief, etc., and each profession has its own unique abilities and characteristics. Players can enhance their abilities by leveling up and acquiring equipment.
3. Game process: Finally, we need to design the game process, including player actions, exploration, combat, tasks, etc. For example, the game can set that players need to explore, solve puzzles, and fight in ancient tombs to complete tasks. During the game, players can encounter various monsters and treasures, and need to constantly upgrade and become stronger to meet greater challenges.
Through the design of the above three aspects, we can create a game based on the rules of the fifth edition of “Dungeon and Dragon”. This game not only allows players to enjoy rich game content, but also allows players to experience real adventure stories.
Finally, do you have any questions for ChatGPT?
This article is reproduced from: https://finance.sina.com.cn/tech/csj/2022-12-06/doc-imqqsmrp8780113.shtml
This site is only for collection, and the copyright belongs to the original author.