In the recent years rise of public usage of artificial intelligence and its products, the question, of whether artificial intelligence will replace and take over peoples work in the future has been raised more. ChatGPT is one of these products of artificial intellignece and machine learning. ChatGPT is a conversational robot, trained on data from the internet up until March 2023 and first became available to the public in November of 2022. ChatGPT's web aplication became the web application that reached 1 million users the fastest and with that worldwide popularity. Soon after ChatGPT's public release users figured out the potential usages of ChatGPT and began using it to replace work. This is how we came to our research that we will present in this thesis, where we checked the ability of ChatGPT to generate articles about NBA games, based solely on statistical data. We also tested the influence of variables within the commands given to ChatGPT for the generation of articles and how those generated articles compare to each other and to real articles. We defined two different commands, and in each we changed the value of a different variable. We named these two variables 'variable of the author', which communicates to ChatGPT the role of the journalist who writes the article, and the 'variable of the employer' of the author, which communicates to ChatGPT on which website the article will be published on. We compared the generated with multiple comparisons, such as clustering, Vader sentiment analysis, LDA, Flesch-Kincaid readability test and cosine similarity, from which we obtained the results that indicate that ChatGPT is currently unable to generate articles similar to those written by professional authors. The difference between the variables themselves, showed us a smaller impact than we expected, but still showed that the variable of the employer generated better articles, compared to the variable of the author.
|
