The constantly developing technology allows many elements that we see in science fiction movies to enter our lives and reveals that a very interesting future awaits us. Artificial intelligence systems, which are advancing gradually, are one of these technologies that are expected to take an important place in our lives. The USA-based artificial intelligence company OpenAI is also among the leading companies in this field.
The technology giant, which has been operating since 2015, had an artificial intelligence system called DALL-E that can transform written definitions into visuals. The company introduced DALL-E2, a more advanced version of this system, in the past months. It was reported that this system has improved features such as having higher resolution, editing images and many more. Now, very interesting information about the DALL-E2 has emerged.
DALL-E2 may have created its own language
A PhD student at the University of Texas in the USA, OpenAI’s DALL capable of converting simple text into photorealistic images He discovered something about the artificial intelligence system called E2. This person named Giannis Daras stated in a Twitter post he shared that DALL-E2 produces a secret ‘language’ and has unique words.
Daras asked the algorithm to generate an image of ‘farmers talking about vegetables’. After that, it was noticed that the algorithm created an image with texts that had no meaning in the resulting image. However, Daras, who examined the system, saw that the word ‘vicootes’ in the resulting image was used for ‘vegetables’ in previously created images. This revealed that artificial intelligence may be using this word in response to the word vegetable.
In addition, Darras stated that the image of the farmers also contained the meaningless words ‘apoploe vesrreatars’. As a result of the searches made in the system, it was seen that these words matched the bird images. It was also stated that the words ‘Contarra ccetnxniams luryca tanniounons’, which were seen to be used by the algorithm, were also matched with the ‘insect’ images.
As a result, theorized that DALL-E2 may have created its own ‘secret’ language. Of course, it is not possible to say that Daras’ results are 100%; because there is a possibility that the algorithm will generate different images. Benjamin Hilton, an analyst, also stated that this situation may have come about by chance; he thinks this is more likely than the system is capable of producing language. Nevertheless, it would not be wrong to state that such a possibility is exciting.
4091
