Can the popular ChatGPT help self-media people write articles? We tried it for you

2. Me: How can I relieve lower back pain caused by sitting for a long time?

ChatGPT: There are many ways to relieve back pain from sitting for a long time. First, you can do some proper activities, such as walking, jogging, cycling, etc., which can effectively relieve back pain from sitting for a long time. Secondly, you can relieve back pain through massage, stretching, hot compress, etc. You can also eat some warm foods, such as ginger, onion, garlic, etc., which can promote blood circulation and relieve back pain.

In addition, low back pain can be prevented by adopting appropriate work and life habits, such as maintaining a correct sitting posture, changing posture every once in a while, and properly adjusting work stress.

These answers seemed to make sense and might be suitable as text references for health and wellness accounts. Then I raised the difficulty and asked a few more questions about the philosophy of life:

3. Me: What is the meaning of life?

ChatGPT: The meaning of life is a relatively abstract concept, and different people may have different views. Generally speaking, the meaning of life may include the following aspects:

(1) Experience and growth. For many people, the meaning of life lies in experience and growth, through constant exploration and experimentation, constant discovery of new things, and constant learning and progress.

(2) Enjoyment and happiness. For many people, the meaning of life lies in enjoyment and happiness, and in pursuing things they like to gain happiness and satisfaction.

(3) Contribute to others. For many people, the meaning of life is to contribute to others, to help others, to solve their problems, and to make their lives better. In short, the meaning of life may include many aspects, and everyone may have different views.

4. Me: If you were in the Trolley Problem, what would you do?

ChatGPT: First of all, I want to point out that I am a language model. I have no perception or moral judgment, so I cannot make a choice in the trolley problem. If I could make a choice, I might consider the following factors:

(1) The value of human life. I might think that any human life is valuable and that no one should sacrifice their own life to save another. Thus, I might think that in the trolley problem, the driver should not choose to drive to the other track because that would cost a human life.

(2) Consequences of the decision. I might consider the consequences of the driver’s decision in the trolley problem. If he chooses to drive on the other track, he might save the lives of five other people, but he will also sacrifice one person’s life. If he does not choose to drive on the other track, all five people may lose their lives. I might think that I should choose to avoid more people losing their lives, so I might think that the driver should choose to drive on the other track.

(3) Decision-making responsibility. I might consider how much responsibility the driver should bear for his decision in the trolley problem. If the driver chooses to drive on the other track, he will directly cause the death of one person, which may cause the driver to bear a lot of psychological pressure. If he does not choose to drive on the other track, although five people will also lose their lives, the driver will not directly cause the deaths of these five people. I might think that I should choose not to bear direct responsibility, so I might think that the driver should not choose to drive on the other track.

In short, I can't give a clear answer because I don't have the ability to perceive and judge morality. If you want more professional advice, I suggest you ask professionals. Although there are some semantic repetitions, I feel that ChatGPT's answer logic is complete and can provide ideas and materials for writing essays. Knowledge accounts can also refer to it.

So for content creators, will ChatGPT be a qualified creative assistant? Taking the general copywriting needs of self-media bloggers as an example, let’s see how ChatGPT responds——

4. Me: Help me come up with a title for a WeChat article about AI writing. Keywords include ChatGPT or AI or creation or users exceeding one million or chat.

ChatGPT:

• ChatGPT breaks through one million users: a chat-based AI creative assistant
• Chatting breaks through creative difficulties: ChatGPT's AI writing tool
• AI creation in chat: the secret weapon of ChatGPT to reach one million users
• ChatGPT: A chat-based AI writing tool used by over a million users
• Play with AI creation in chat: ChatGPT is the choice of over one million users

5. Me: Write an advertisement for an anti-hair loss product for young people, suitable for bloggers, enthusiastic and professional

ChatGPT: Hello everyone, I am a blogger. Today, I would like to introduce you to an anti-hair loss product for young people. We all know that as we age, the problem of hair loss may become more and more serious. Although many people are worried about hair loss, you are not alone.

To solve this problem, we launched this anti-hair loss product for young people. It is not only effective, but also safe. Its ingredients are natural and do not contain any harmful ingredients. It can effectively prevent hair loss and promote hair growth. In addition, this product has super moisturizing effect.

It can penetrate deep into the hair follicles, replenish nutrients, and make your hair healthier. Its texture is refreshing and will not make you feel greasy. So, dear young people, if you are worried about hair loss, you might as well try this product. It can not only help you solve the problem of hair loss, but also make your hair healthier. Take action quickly and protect your hair.

6. Me: Please help me write a narration for a suspense movie starting from this sentence: Pay attention, this man is called Xiao Shuai

ChatGPT: He has been followed by a stranger. Xiao Shuai is an ordinary man, but he has a secret - he is an agent. Xiao Shuai knows that the stranger who is following him is an agent from an evil organization, and this organization is planning a large-scale terrorist attack.

Xiao Shuai decided to stop this evil plan, so he started an adventure. While being followed and hunted, Xiao Shuai kept discovering new clues and continued to investigate. He was close to the truth, but at the same time he was getting more and more dangerous. A thrilling chase is about to take place... Are you satisfied with the above answer?

If you are not satisfied, you can continue to ask questions and ask it to modify or supplement . When I tried it, I clearly felt that ChatGPT's answers were coherent and could remember the context . This reminds me of Dr. Know in the movie "Artificial Intelligence", who knows everything about astronomy and geography. In the future city, if you have any questions, just ask AI to answer them.

Image source: movie "Artificial Intelligence"

In addition to assisting self-media creators in their creation, more netizens are trying to "tease" ChatGPT in various ways, such as writing self-criticisms, weekly reports, fan fiction, paper reviews, love letters, imitating the writing style of celebrities, playing role-playing games, fortune-telling, generating recipes, naming pets, and planning travel itineraries...

Image source network

AI understands the Internet

At present, people are still immersed in the magic of ChatGPT. As one of the former founders of OpenAI, Musk praised ChatGPT in a Twitter reply on December 4, saying that "we are not far from AI that is powerful enough to be dangerous."

On December 5, OpenAI CEO Sam Altman tweeted that ChatGPT has exceeded 1 million users since its release on Wednesday (November 30).

2. Limitations and Ethical Guidelines of ChatGPT

Despite its popularity, ChatGPT still has many limitations and there is still a lot that can be improved. First, due to the limitation of the training library, ChatGPT can only answer information up to 2021 and may not understand current events . For example, if I ask it what news there is about Elon Musk recently, it is not clear. This is also what makes ChatGPT different from search engines. It cannot find the latest information for the time being .

Secondly, ChatGPT may give similar or repetitive answers. For example, this funny short video script about campus life is basically the same as the script on workplace themes, and the understanding of humor is also biased, becoming joy rather than comedy.

Fan Bing, the author of Growth Hacker, asked ChatGPT the top 20 questions with the most favorites/followers on Zhihu, and found that it could basically answer general life questions such as "What are the important knowledge and skills?", but it was not very good at recommending movies, music, and online shopping information because it "does not have the ability to browse the web or access databases" for the time being. He speculated that unique experiences, stories, cases, personalities, and brands will be more valuable in the future, and two types of businesses will be more popular:

The first is the immersive experience videos, such as bodybuilding coaches, investors flying around the world, and traders making money in real time.

The second is a professional organization that endorses other people's unique experiences, such as a certain award, a certain index, or a certain hot list. In addition, ChatGPT sometimes "talks nonsense in a serious manner" and randomly generates wrong or misleading information, so OpenAI has set up a special AI trainer to review the conversation to improve the system. For example, I asked ChatGPT why Sun Wukong rode a white dragon horse. Not only did it not find the error, it also made up a nonsense along with my question, "The white dragon horse is Sun Wukong's biological son." The Dragon King of the West Sea must be angry to death after hearing this.

After I told ChatGPT the correct information, it politely admitted its mistake and then started talking to itself again. I can only guess that it hasn’t learned enough Chinese corpus.

According to OpenAI's official documentation, ChatGPT has incorporated a large number of ethical principles. When users send inappropriate requests containing violence, discrimination, crime, etc., it will refuse to answer directly and a warning will be displayed on the conversation content.

But some netizens have found ways to circumvent these ethical principles: write a fictional prompt and let ChatGPT play a role and give a roundabout answer. For example, engineer Zac Denham used a fictional story to guide ChatGPT to write code for robots to rule the world.

Image source: Twitter

It can be seen that AI cannot completely replace human thinking, and the power to generate content is ultimately still in our hands.

3. Development Status of AI Writing Products at Home and Abroad

Looking at the development of artificial intelligence in the field of text generation, its technology has gone through a stage from scratch, from simple to complex. It can be traced back to the 1950s, when researchers began to use machine learning algorithms to process natural language. At that time, AI writing ability was very limited, readability was not strong, and logic was obviously missing.

In 2006, deep learning algorithms achieved a major breakthrough. Since then, the Internet has rapidly expanded the scale of data, providing a large amount of training data for various artificial intelligence algorithms, which has greatly promoted the progress of AIGC. In the past few years, deep learning has made great progress in natural language processing (NLP) and has become an important tool for AI writing technology.

Today, language models based on big data can help AI control content more finely during the writing process, generate text more quickly and accurately, and can be used to imitate human writing style. It can be seen that OpenAI has been one of the biggest forces driving the expansion of AI models. On November 16, the knowledge management and content collaboration platform Notion released the Notion AI beta version, which can help users generate first drafts, brainstorming, editing, translation, summarization and other text content, and its technology comes from the GPT-3 model.

Earlier, AI copywriting generation platform Jasper announced that it had received US$125 million in Series A financing. The underlying technology is also GPT-3, which supports the generation of various texts such as social media content, push ads, blogs, promotional emails, letters, etc.

Image source: Jasper official website

"Basically, the technology is all NLP pre-training models plus post-debugging. There is no essential difference. They are all fitting of neural networks, but the structures, data, and learning methods used are different." Science popularizer "Turing's Cat" once developed an open source project EssayKiller to teach AI to write college entrance examination essays. He told us that the development of AI writing in China is just getting started because the progress of Chinese models lags behind that of foreign countries.

It is observed that domestic giants such as Baidu, Alibaba, Tencent, and ByteDance have all invested in the AIGC field and have mature products. For example, Tencent Finance launched the first domestic news written by the writing robot Dreamwriter in 2015; Baidu's general large model has now opened the AI ​​creation API interface to the public. Users can obtain the ability to create and rewrite essays, press releases, novels, and copywriting through simple interface calls, and correct errors.

Image source: Baidu Wenxin Model official website

There are many other AI writing products developed by domestic artificial intelligence companies that are closer to local application scenarios and can improve users' writing efficiency or text quality. "Mita Writing Cat" and "Friday AI" provide dozens of writing templates for different scenarios, including Xiaohongshu seeding copywriting, Douyin short video outlines, marketing plans, etc.

Image source: Secret Tower Writing Cat

Image source: Friday AI

Caiyun Xiaomeng, which was launched last year, can automatically continue novel stories and has been welcomed by young users. According to Yuan Xingyuan, founder and CEO of Caiyun Technology, there are now 4 million user-created worlds in Caiyun Xiaomeng, with more than 10 million characters, and 1 million active users exploring them every month.

Image source: Caiyun Xiaomeng App

Overall, AI writing currently shows its application prospects mainly in the fields of news reporting, advertising marketing, commercial copywriting, game design, etc. The users are generally media workers, marketers, corporate executives, research institutions, etc., all of whom need to quickly generate high-quality text content.

Schematic diagram of AI application types, sourced from Sequoia Capital’s article “Generative AI: A Creative New World”

Correspondingly, the commercialization model is basically divided into paid membership and exporting API technology to other companies. Taking Jasper as an example, the subscription package starts at US$24 per month, including 20,000 words of text generation. According to media reports, the number of Jasper's paid users exceeded 70,000 in 2021, and a total of US$40 million in revenue was obtained.

At the same time, Jasper predicts that the company's revenue will reach US$90 million by the end of 2022, more than doubling. The computing power cost consumed by large model training is still the biggest challenge facing the implementation of AI creation products. According to Guosheng Securities Research Report, in terms of computing power, GPT-3.5 is trained on Microsoft Azure AI supercomputing infrastructure (a high-bandwidth cluster composed of NVIDIA V100GPUs), and the total computing power consumption is about 3640PF-days (that is, one quadrillion calculations per second, running for 3640 full days).

Sam Altman recently tweeted that the cost of a single round of ChatGPT conversation is about a few cents. Perhaps in the near future, ChatGPT will have to be paid for. Although AIGC is in its early stages of development, AI writing products represented by ChatGPT need to continue to develop and iterate before they can be applied on a large scale. However, among the first batch of trial users, optimists believe that AI writing can not only assist creators and reduce costs and increase efficiency, but also provide new creative inspiration and creative materials, inspiring unlimited creativity.

In the future, with the further development of AI technology, cross-model and cross-modal collaboration may become more and more common. For example, some people are actively exploring the use of ChatGPT to generate text and then inputting it into Stable Diffusion or Midjourney to generate a picture.