One year later: Scientists are experimenting with ChatGPT

A year after its sensational debut – 180 million people have used it so far – and despite the many controversies that have accompanied this “conversational agent” – from the misinformation it produces to the jobs it threatens – we have found uses for it that are far from being that. Being negative. British review nature Take advantage of this birthday, November 30th, to give the word to six researchers who used it with a certain enthusiasm. One, computer scientist Abiba Birhane, of Trinity College, in Ireland, is in the other camp: she wrote that the “potential benefits” of such “artificial intelligence” were “systematically overstated” and that insufficient thought had been given to its “accuracy and reliability.” its performance and the data on which the model was trained,” or whether the model code itself should be openly accessible.

As for the other six, even though they used it, they did not transform. Science communication researcher Siddharth Kankaria of India’s National Center for Biosciences writes that initially, “I was excited by the promise of ChatGPT for communicating science. It seemed like it could write scientific research summaries that were clear, eye-catching, accessible, and helped simplify terminology.” Today, he finds that many These applications “often need a lot of attention and checking,” which led to him using these tools in “more targeted ways.”

Like others before him, Ethan Mollick of the University of Pennsylvania has found that teachers will need to rethink homework as a means of assessment. On the one hand, because the text “detectors” generated by AI will remain imperfect, and on the other hand, because the ability of AI to answer students’ questions will improve further. On the other hand, classrooms will continue to provide added value: “opportunities to practice learned skills, collaborate to solve problems, support from teachers and socialization.” The challenge facing education is not without risks, but artificial intelligence, Ethan Mollick believes, could enable experimentation with new ways of training the next generation of students.