TU Delft - Feeding ChatGPT with more than 2000 abstracts to predict scientific impact

In previous posts, I emphasized the importance of the metric Mendeley readership, highlighting its capacity to forecast early scientific influence (this capability holds particular significance for young researchers), namely in a post from August 2020 https://pacheco-torgal.blogspot.com/2020/08/an-h-index-for-mendeley-comparison-of.html and in another post from January 2021, about a study that confirmed the importance of the aforementioned metric, —though regrettably, the authors of the study overlooked citing important articles https://pacheco-torgal.blogspot.com/2021/01/early-indicators-of-scientific-impact.html and finally in another post from October 2021 about the (German) excellence mapping tool https://pacheco-torgal.blogspot.com/2021/10/paper-excellencemappingnet-new-release.html

In the aforementioned context, I consider it imperative to highlight a recent and innovative study conducted by a professor of Cognitive Human-Robot Interaction, Joost de Winter, affiliated with TU Delft University. This study used ChatGPT-4 (the paid version) in conjunction with a comprehensive dataset consisting of more than two thousand abstracts to predict scientific impact.

Update September 27th - A search carried out today on Scopus reveals that there are 3476 publications with the word ChatGPT. The ratio of publications per million inhabitants for the USA, UK, Germany, and Switzerland is presented below. It is also presented for each of these countries the top 3 affiliations:

Switzerland......9,5 pub./pop.

Top affiliations__Univ. Zürich, Univ. of Bern, ETH Zürich

UK...................5,3 pub./pop.

Top affiliations__Univ. College London, Univ. of Cambridge, Univ. of Oxford

USA................3,2 pub./pop.

Top affiliations__Stanford Univ., Harvard Medical School, Duke Univ.

Germany.........2,7 pub/pop.

Top affiliations__Tech. Univ. München, Ludwig-Maximilians-Univ., Friedrich-Schiller-Univ. Jena