The GPT-2 artificial intelligence program is capable of writing fake news articles. At the time of the fake news, it represents a danger, and its designers have decided not to make it public.
The " fake news " has become the haunt of democracies by their virality on social networks, and draft laws appear even to counter this phenomenon so dangerous, able to manipulate the masses. It is for this reason that GPT-2 will remain confidential and will not be made available to the public and businesses.
Developed by OpenAI, a non-profit artificial intelligence research company supported by Elon Musk, it is a text generator. But it's not just simple texts, but reports, press articles, and even fiction! So far, that's fine ... except that the machine is so powerful that it is able to create " deep fakes for the text ", more than credible false information created by artificial intelligence, and even the share it yourself on social networks!
With structure and style
It is precisely because their machine has managed to create crying articles of truth that the researchers have decided to keep it in the boxes, and not to share it. Initially, the artificial intelligence system collects texts, and it varies from a few words to a dozen lines. Then, the researchers propose to their program to generate the continuation or to create their own article. And this is where the machine has bluffed them because it is able to keep the style of a text, but also to structure it, thus erasing the automatic character and word by word of the first text generators.As a result, researchers have chosen to curb the development of the machine to prevent it from falling into the hands of malicious people. " We have to experiment to find out what they can and can not do, " says Jack Clark at the Guardian about the malicious use of GPT-2. " If we can not anticipate all the possibilities of a model, we must be able to see what it can do. There are people who are much more aware of what we can do with them. "
No comments:
Post a Comment