Do you remember the artificial intelligence dubbed GPT-2, specialized in the generation of realistic texts and whose developers (OpenAI), in an unprecedented decision (and quite criticized), decided not to publish it, because of the danger they felt that conferred his ability to spread fake news?
That was last February. Well, now (9 months later), OpenAI has decided that the lion was not as fierce as they painted it, or – in other words – that “there is no strong evidence of misuse” of the more incomplete versions that have been released in these months, so they have proceeded to release the entire model.
GPT-2 is based on an evolved natural language processing technique known as ‘language models’, which become machine learning models dedicated to predicting what the next word in a text should be by virtue of all the above words.
One of the peculiarities of GPT-2 is that, from an initial text (provided by us), it is able not only to continue it to generate a broader text, but can also be configured to perform translations or summaries of it, and even to answer questions about its content.
When our goal is to generate fake news, we may encounter very convincing texts that may seem the product of an intelligence, by the way, the phrases and themes are spun, but the little bit we play with GPT-2 long enough start to make clear the limitations of the model.
Maybe it wasn’t that big to deal…
One of its great flaws, for example, is long-term coherence: the names and characteristics of the characters mentioned in the text may end up varying throughout the text. There have also been cases where GPT-2 has generated texts that speak about the “four horns” or about “underwater fires”. You can check it yourself using GPT-2 through a web interface like TalkToTransformer.com.
Jack Clark, director of Policies of OpenAI, explained in February that the reason that led to not publishing the full version of GPT-2 already in February was not only that it could be used to generate very convincing fake news, but that It would facilitate its automation and optimization (based on demographic factors) for specific social sectors.
Nvidia’s research director, Anima Anandkumar, then fiercely criticized OpenAI’s decision not to make the model public:
“Where is there any evidence that your system is capable of doing that [what do you say it does]? What independent researchers have analyzed its system? No one. If you think it really is capable of that, you will open it to researchers, not to the media that are looking forward to the clickbait. “
Anyway, just in case there is some truth in OpenAI’s suspicions, they have continued to investigate in the field of automated systems, not so much to write new ‘fake news’ as to help detect texts that have been created using GPT-2.