How and why we use ChatGPT, and with what risks

A survey of a thousand US users reveals the chatbot’s browsing habits. Not all employers are in favour.

Posted on April 11, 2023 by Valentina Bernocco

ChatGPT is a flash in the pan or the beginning of a true technological, cultural and maybe even anthropological revolution? The symbolic application (or at least the most famous) of generative artificial intelligence made "popular" is currently not accessible in Italy, waiting for OpenAI to clarify the doubts of the Data Protection Authority. In the rest of the world, however, users continue to use ChatGPT and a survey conducted in the United States shows how and why. The study, carried out by the website WordFinder, analyzed the behavior of 1,024 US users and 103 AI experts, 57% men, 39% women and 4% identified as nonbinary. More significant, for statistical purposes, is perhaps the fact that Millennials are by far the most represented generation in the sample (62%), followed by Gen X (22%), while the personal details of Baby Boomer (7%) and Gen Z (9%).

The study, however, provides an interesting insight. 46% of respondents said they used ChatGPT only once or twice, and therefore it falls into the category of the simple curious, but there are regular users: 10% turn to the chatbot several times in the course of a month, 20% once a week, 19% several times a week and 5% every day. It should be said that in all age groups the share of those who use ChatGPT regularly (several times a month or more) is over 50%.

Employers for or against?

Even more interesting is the fact that many consult artificial intelligence secretly. One in four employees used ChatGPT for professional reasons and 29% did so without informing their employers (the percentage is higher among those involved in business, while marketing tends to use the chatbot in transparency). Some were discovered and the fact had no consequences in 68% of cases, but there were even higher who wanted more details on how to use the application (18%) and few who scolded the employee (4%) or asked not to repeat the error (5%). In 2% of cases, the use of ChatGPT without permission led to dismissal.

Of course, without knowing the specific context of each episode it is difficult to make a judgment on the severity of the possible consequences, on what is right or wrong to do. What is certain is that the technological advances of AI cannot be ignored, like it or not.

(Image of macrovector from Freepik; opening image taken from Freepik)

Why and how to use ChatGPT

The tool developed by OpenAI in the United States is used at work mainly to find or invent ideas (41% of respondents), to create content (20%), to respond to emails (14%), to write software code (11%)to write cover letters and resumes (10%) and to create presentations (9%).

In the context of work uses, 50% of respondents write prompts by themselves (ie the text strings that give instructions to the program), others search for them on YouTube (14%), OpenAI resources (12%), ask friends (11%), search for them on TikTok (10%), Twitter (9%), Facebook (8%), Instagram (5%) or in some cases ask work colleagues for help (4%).

ChatGPT is also used as a source of information and knowledge, a bit like we do for decades with search engines, or as a "help" to write texts or to study. In particular, 33% of respondents use it to find information on a certain topic, 18% to help them understand concepts that are difficult to understand, 15% to learn new skills, 11% to write texts, 6% to get the summary of a book, 4% to get answers for homework and only 3% use it to study for an exam.

Concerns and reservations

In recent weeks ChatGPT has been at the center of several concerns made explicit. As the United States started a petition with thousands of petitioners (including Elon Musk) calling for a pause in the development of evolutionary generative AI models, in Italy, the Italian Data Protection Authority challenged OpenAI’s lack of transparency on data collection and insufficient protection measures for minors.

There are also fears about the impact on employment. A recent analysis by Goldman Sachs estimates that generative AI technologies can cause the loss of 300 million jobs in the medium term, replacing about 18% of current employment. On a smaller scale, the aforementioned WordFinder study also gives some indication: 6% of respondents fear that they may lose their jobs because of ChatGPT. A small percentage after all.

A few numbers on ChatGPT

Until yesterday, machine learning models were able to perform the specific tasks for which they had been trained, for example image classification, trend prediction from a database, text analysis. Today, however, the models behind generative AI applications, namely Foundation Models, can be used to create different applications, with different or multiple abilities, ranging from processing responses to writing texts, from the resolution of mathematical problems to the generation of images, up to specific activities that meet the needs of the individual company.

The evolution of these technologies has been rapid and surprising. In 2019, the larger pre-allected model contained 330 million parameters, while GPT-3.5 (the large language model behind the current public version of ChatGPT) now has 175 billion parameters. And it is not known, but it is certainly much higher, the number of parameters included by OpenAI in the latest evolution, GPT-4.

Introduced last November, the public version of ChatGPT is based on GPT-3.5, the 3.5 generation of the language model (large language model) developed by OpenAI. The acronym GPT stands for Generative Pre-trained Transformer, indicating that the program knows how to manipulate language and create content based on a "training" already completed. Over 45 terabytes of data from books, academic publications, and the entire Wikipedia knowledge base have been used to train GPT-3. In total, the model has been trained on 175 billion parameters, a number that impresses but is still far from the 100 trillion parameters average processed by the human brain.

The most advanced model, GPT-4, is described by OpenAI as superior to GPT.3.5 in that it is "more reliable, more creative and more capable of handling instructions with nuances of meaning". Moreover, it is not a simple linguistic model but a multimodal model, for which texts and images were used for training. OpenAI has not revealed the number of parameters managed by GPT-4, but many have speculated that it could approach 100 trillion of the human brain. Currently to access the program you need to get on the waiting list, knowing that priority will be given to software developers.

Meanwhile, the free version of ChatGPT has beaten the fastest-growing application in the history of computer science: last November in the first week since launch registered a million users, while in January the ChatGPT web application had an average of 13 million daily unique visitors.

The statistics of the standalone application do not, in any case, cover the whole phenomenon. Software giants have begun to integrate generative AI into their products, starting of course with Microsoft (the main OpenAI investor), which has brought it into the Bing search engine, the Dynamics 365 suite and some Azure services. The Redmond company is also working to integrate ChatGPT into Microsoft 365 applications, including Word, Excel, PowerPoint, Outlook. Alphabet is pursuing a similar strategy with Bard, its language model, gradually integrating it into Google Search.

Tags: work, artificial intelligence, openai, chatbot, chatGPT, Ai generativa

Leave a Reply

Your email address will not be published. Required fields are marked *