Four questions about ChatGPT answered

The software tool ChatGPT from the company OpenAI can already write entire pieces of text. Should we worry about the technology?

ChatGPT has become a real hype. In late November, OpenAI, an artificial intelligence research lab, launched the chatbot. Since then, many people have used it. Does this mean that those with a writing major can put down their pen? KIJK answers four questions about ChatGPT and lets the chatbot answer the questions itself.

Also read:

1) How does ChatGPT work?

SEE: The chatgenerative pre-trained transformer (ChatGPT) is a so-called Large Language Model, a complex AI language model that helps computers respond to human language. The chatbot was created using two learning techniques: watched and reinforcement learning. Bee supervised learning, or supervised learning, an algorithm learns to see relationships using examples. The chatbot is shown an example of the input and the desired output. The goal is then to learn a general rule that translates the given input into that output. supervised learning is used, for example, to classify unwanted e-mails as spam and put them in a separate folder. Bee reinforcement learning the algorithm learns by trial and error (trial and error). Desirable behavior earns extra points, mistakes do not.

Verbatim response from ChatGPT: ChatGPT is a conversational language model developed by OpenAI. It is based on the Generative Pre-training Transformer (GPT) architecture, which was introduced in 2018 by Google Brain. The GPT model is pretrained on a large amount of text data and can be tuned for various language processing tasks, such as language translation, text summarization, and question answering.

ChatGPT is a variant of GPT designed specifically for conversational AI applications. It is trained to generate human responses to text input in a conversational context. ChatGPT is trained on a diverse set of conversational data, such as online conversations, customer service chats, and movie scripts, to provide a broad understanding of how people communicate.

With the ability to generate human text, ChatGPT can be used to build chatbots and virtual assistants that can have natural conversations with humans. This also applies to creative writing.

As a variant of GPT-3, the model is very powerful and can be tuned for a wide range of conversational AI tasks.

2) What can the language model do?

SEE: As ChatGPT’s input in this article shows, it can formulate responses to questions you ask or commands you give (“write a Christmas speech to my colleagues”). But the applications go much further than that. ChatGPT can write poems, full stories and advertising texts. Furthermore, OpenAI foresees a future as a search engine for the chatbot. Currently, the algorithm is not yet live connected to the Internet, but trained with information up to and including 2021. So as a search engine, it is not yet very reliable and up-to-date.

ChatGPT: Here are some examples of the many ways ChatGPT can be used:

  1. Chatbot Development: One of ChatGPT’s most popular uses is building chatbots that can have natural conversations with humans. The model can be tailored to specific domains, such as customer service or e-commerce, and it can generate human responses to text input.
  2. Text Generation: ChatGPT can be used to generate text in various formats, such as articles, stories and poems. Because the model is trained on a diverse set of text data, it has a broad understanding of how language is used, enabling it to generate coherent and believable text.
  3. Text Completion: ChatGPT can be used to complete passages of text, be it a sentence, paragraph or even an entire article. It works by using the context of the given text to generate the most likely continuation of the input.
  4. Question Answering: ChatGPT can be tailored to answer questions based on a given context or specific knowledge. This can be useful in specific domains such as customer service, education or research.
  5. Language Translation: Through fine-tuning, ChatGPT can be used to translate text from one language to another, and it can also be used for text summarization.
  6. Dialogue Generation: ChatGPT can be used to generate dialogue between multiple characters in a story, script or conversation. This can be useful for creative writing or dialog generation in video games, simulations, and other interactive applications.

These are just a few examples of the many ways ChatGPT can be used. The model is very flexible and can be set for a wide range of language processing tasks.

3) What can’t the chatbot do (yet)?

SEE: ChatGPT is trained with all kinds of data. So the answers the algorithm gives are based on existing information. This means that it cannot gain new insight on its own. “It has only been trained in linguistic fluency and to produce texts that fit what the user wants, but not truth and correctness,” said Jelle Zuidema, associate professor of natural language processing at the University of Amsterdam. de Volkskrant.

Another major drawback of ChatGPT is how much energy the servers running the language model consume. Sam Altman, CEO of OpenAI, has therefore already indicated that the company will one day charge money to use the chatbot. But that’s not even the worst part; because these servers use so much energy, training ChatGPT is accompanied by a high CO2emissions.

Chat GPT: ChatGPT is a very powerful conversational language model, but like any machine learning model it has limitations. Here are some things that ChatGPT cannot account for:

  1. Understand context or intent: ChatGPT can generate human responses, but it doesn’t really understand the context or intent behind the text input. It is unable to infer the underlying meaning or feeling of the input, so it can only respond based on the specific words and phrases it saw during training.
  2. Solve problems that are too complex: Although ChatGPT can execute a wide variety of language processing commands, it is unable to solve problems that are too complex or abstract. For example, it would have difficulty understanding complex scientific or mathematical concepts.
  3. Recognizing emotion, sarcasm, or irony: The model may have difficulty understanding the nuances of human emotion, sarcasm, irony, and so on. So it may not provide an appropriate response to input with such elements.
  4. Judging the correctness of the output: ChatGPT is a language model, it is unable to judge the correctness of the output, it can only generate text based on the input, but it does not guarantee that the output is correct or meaningful.
  5. Decision Making: ChatGPT has no ability to make decisions based on input, it can only generate text based on the input it receives.
  6. Learn from conversation: Although ChatGPT can respond in a conversational context, it does not learn during the conversation. It relies on a pre-trained model with a specific context, and it cannot adapt to new context or new information during the conversation.

Overall, ChatGPT is a powerful language processing tool, but it has its limitations. It is important to understand its capabilities and limitations if you decide to use it for a particular task.

4) Should people with a writing major fear their jobs?

SEE: Certainly not. As explained above, the chatbot is trained with information up to and including 2021. This ensures, firstly, that the given output is not completely up-to-date, in addition, the output is not checked for factual accuracy and does not possibly contain new insight. Last week, Ilyaz Nasrullah, digital strategy consultant and columnist, registered Fidelity that ChatGPT is “very stupid and completely unreliable”. Artificial intelligence will therefore certainly not replace journalists, writers or lawyers – to name a few examples.

However, the algorithm can be a useful tool for some professions. Magazine The engineer cites computer programmers as an example. “They spend a good part of their time figuring out how to formulate a function in a particular programming language. A software tool can help with that,” the website says.

Chat GPT: ChatGPT and other language models can be powerful tools for generating high-quality text, but they should not be seen as a threat to jobs in the writing or journalism industry.

Language models such as ChatGPT can be used to improve the work of human writers by providing inspiration, suggestions for sentence structure, or even entire paragraphs of text in certain situations. However, it is unable to replace the human judgment, creativity and emotion that a human writer can put into a piece of text.

Language models can also be used to automate routine writing tasks, such as writing summaries, news articles, and other types of content. However, this does not mean that the job of the human writer is completely replaced, instead it can lead to more time and resources to focus on more complex and creative writing tasks.

The real potential of language models like ChatGPT is to help and enable human authors to produce higher quality and more efficient work, while opening up new opportunities and allowing them to focus on higher value-added tasks.

It is also important to mention that although ChatGPT is a powerful model, it is still highly dependent on the data it has been trained on, it may not have the knowledge, cultural understanding and context of certain topics, especially those that are niche or specific. So the role of human authors in those cases will continue to be important.

Overall, language models like ChatGPT have the potential to change the way we write and communicate, but they won’t completely replace human writers. They will complement and enhance rather than replace the work of human authors.

Sources: ChatGPT, Trouw, De Ingenieur, de Volkskrant

Image: OpenAI

Leave a Comment