Sponsored by Elon Musk non-profit research organization OpenAI told The Guardian about his new development, a revolutionary AI system that knows how to write a news story and literary works. That’s just the developers decided not to publish scientific work on the AI: they are afraid that the development could use with malicious intent.
Relying only on a couple of lines of the source text, AI called GPT2 know how to append text in such a way that it will be stylistically and thematically match the original sample.
The Guardian journalist was able to play with the development. Give the AI a couple of paragraphs of his article about Brexit, the journalist received quite plausible text, which even had “quotes” politicians, mention of the Irish border. Of course, the text was plausible, but was not different with truthfulness.
AI was trained on 10 million articles, links to which were obtained from the website Reddit. Overall gained 40 GB of text equivalent to 35 000 copies of the book “Moby dick”.
OpenAI as a nonprofit organization talks openly about his research. But in this case, the creators of artificial intelligence fear that it can be used by hackers, spammers and creators of false news.