A research organization specializing in artificial intelligence has announced that the automatic text generator it has developed will remain confidential, worried that its program with very good results could fall into the wrong hands. The program, dubbed GPT-2, can be used for articles, presentations or any writing. This is ” the ultimate for modeling certain language levels ”, including the summary and translation, according to the research center OpenAI, which is backed by Tesla boss Elon Musk, as well as Amazon and Microsoft.
But this revolutionary program will not be revealed to the public because ” of our concerns about the malicious uses of this technology ”, OpenAI researchers explained on their blog on Thursday. According to them, the program could be used to generate fake news articles, impersonate other people online or automate infoxes on social networks. The program thus wrote a 300-word article based on the following premise: ” a herd of unicorns living in a valley of the Andes which had never been explored so far ”. OpenAI ensures that the GPT-2 ” is better than other language models ” after being tested with Wikipedia, information or books, without needing specific training.
” The general public must be more suspicious of texts published online, as the phenomenon of ‘deepfakes’ must make people more suspicious of images ”, noted the researchers, with reference to manipulated videos that reproduce a person’s body language and tone of voice. Artificial intelligence is the science of which the goal is to make a machine perform tasks that man performs using his intelligence and cognitive abilities. The American firm IBM was a pioneer in this field with the victory of its supercomputer Watson against a great chess player or that during the general culture game Jeopardy. However, its “Project Debater” program recently failed to convince an audience during a debate on a social issue organized against a human.