Can ChatGPT replace the knowledge factor?

OpenAI’s new chatbot has impressed many with its ability to write scripts and code. What are the risks?

A new chatbot from OpenAI has revolutionized the world. In less than one month, over 1 million users have signed up, and the service can be used for everything from School style cheating To write and publish books weekend. Will it make knowledge workers redundant? Have we opened Pandora’s Box?

British economist John Maynard Keynes predicted in 1930 that the workweek would be reduced to 15 hours due to advances in technology. Now, nearly 100 years later, OpenAI has launched chat, a chatbot that can communicate naturally with humans and solve a variety of tasks. But what does this mean for knowledge workers? Will ChatGPT take over their jobs? Well, you might want to take a look at the story. Machines have always helped people do things faster and more efficiently, but they have also opened up new opportunities and created new jobs. ChatGPT will probably do the same thing. Instead of replacing humans, it can help them solve tasks more efficiently, allowing them to focus on more creative and meaningful tasks. So Keynes’ prediction of a 15-hour workweek may not immediately become a reality, but at least knowledge workers can rest assured that ChatGPT won’t steal their jobs.

Have you been deceived?

If you thought the previous paragraph was written by us, then ChatGPT has actually succeeded in deceiving you. It was created when we asked the chatbot for feedback about the launch, with reference to Keynes. Then it is not difficult to be affected, but at the same time also to worry.

ChatGPT can carry out clear dialogue in different languages ​​- including Norwegian – and serves as a tireless personal assistant, whether you need help with coding, a promotion or an article on any topic. It’s never been easier: Over a million curious people have used the service to create everything from poems, websites, and text summaries to a Bible Describe On how to remove a peanut butter sandwich stuck to a VCR.

But it didn’t take long before users started pointing out the limitations. Large language models, which we looked at at Telenor, are pretty good at predicting the next word in a sequence, but that’s it. in a professional literature They are called “random parrots”: the models were trained to put together sensible sentences, based on patterns found in training material from the Internet. So it is also not difficult to make the form produce racist statements or advice harmful to health, despite the reduction of content filters. These models will always reflect the source material they have been trained on, and they cannot think analytically.

Even if the answers are well phrased, the content should always be carefully considered. OpenAI itself warns that the chatbot mixes jokes and facts in a convincing manner. Runs, for example, without a problem clarification Benefits of adding china powder to infant formula. This is because the model groups together words that fit the context, without evaluating the truth value of the generated text.

By the way, the chatbot can’t point to related resources, or source material at all. This runs the risk of inadvertently stealing someone’s ideas and committing plagiarism.

Having said that, it is of course exciting to see advances in technology and the potential to simplify the jobs of many types of knowledge workers, such as copywriters and journalists. A usable starting point will save a lot of time, even if everything that ChatGPT outputs should be checked for factual errors and argument errors.

For those of us who are looking for digitization, ChatGPT primarily shows the great potential inherent in the technology. Last week, for example, the last phone call was made over the copper network. The ancient telephone network has revolutionized our daily lives. Now, 142 years after the first chat happened, you can have a chat with ChatGPT and even get some writing help, but the result is unlikely to be a two-day Keynes work week.