Researchers experimenting with OpenAI’s text-to-image instrument, DALL-E 2, observed that it appears to covertly be including phrases reminiscent of “black” and “feminine” to picture prompts, seemingly in an effort to diversify its output


22 July 2022

This photo illustration shows the DALL-E website reflected between a computer screen and cellphone in Washington, DC, on July 20, 2022. - A million people eager to dabble with a new artificial intelligence tool that lets them create images simply by describing them with words will soon get their wish, its creators said Wednesday. Artificial intelligence research firm OpenAI is conducting a wide-scale beta test of DALL-E, a cutting-edge software that creates images from textual descriptions. (Photo by Stefani Reynolds / AFP) (Photo by STEFANI REYNOLDS/AFP via Getty Images)

DALL-E 2 is a text-to-image producing synthetic intelligence

STEFANI REYNOLDS/AFP through Getty Photos

Synthetic intelligence agency OpenAI appears to be covertly modifying requests to DALL-E 2, its superior text-to-image AI, in an try and make it seem that the mannequin is much less racially and gender biased. Customers have found that key phrases reminiscent of “black” or “feminine” are being added to the prompts given to the AI, with out their information.

It’s well-known that AIs can inherit human prejudices by coaching on biased knowledge units, usually gathered by hoovering up knowledge from the web. For …

By 24H

Leave a Reply

Your email address will not be published.