We put MIT technology to the test!We compared, GLTR and Talk To Transformer, two systems based on Artificial Intelligence to complete texts.
Artificial Intelligence is our multi-tasking assistant. One of its functions is to help us improve our writing automatically with programs like Microsoft Ideas. The ability of AI to create texts, in an era where deepfakes are increasingly worrying users, has led MIT to create a tool that detects these automatic texts.
Giant Language Model Test Room (GLTR) is the tool capable of detecting which parts of a text are created by an algorithm.
Words are identified by color. When a word is most likely to appear after the previous word, it appears green. When less likely, they are marked in yellow and red, and finally in purple. If a text has many yellow, red, and purple labels, we can suspect that it was generated with an AI.
GLTR will allow to discover the fake news that are created with Artificial Intelligences
While tags are very useful for analyzing text in more depth and finding out if it has been written by an AI, word detection alone, it would not be enough to state whether a text has been written by a human or not.
An example of AI that could pass GLTR controls is Talk To Transformer. Application generates entire texts based on phrases that we indicate. Its main limitation is that, for the moment, it is only able to interpret the text if we write it in english.
We have decided to try Talk To Transformer, indicating only one word: BlogThinkBig. And this is what we’ve got:
BlogThinkBig has been a leader in promoting quality leadership development by offering expert advice on managing professional development, mentoring, consulting and mentoring. We have been a leader in developing and connecting leaders of different disciplines in the field of leadership to enhance the careers and lives of their colleagues. Our staff have over 30 years of experience in mentoring and education with leadership, policy and development professionals of all degrees in the US and abroad, and currently include many professional leaders of all disciplines.
We are open to new perspectives from the diverse range of leading leaders and experts, and welcome expert reports and peer-reviewed articles.
After generating the text, we analyzed it in Giant Language Model Test Room and this has been the result:
Test in Giant Language Model Test Room GLTR
It seems that so far GLTR has not managed to discover Talk To Transformer, although it is important to remember that GLRT is still in the demonstration phase.
Dare to try Talk To Transformer and Giant Language Model Test Room!
If you are interested in AI do not miss how Aura, the AI of Movistar Home works.