One of the new features included in DALL-E 3, the third version of OpenAI‘s image generator, was the complete censorship of controversial symbols and public figures. Nevertheless, users have managed to bypass all protections, and there are already those who recreate scenes where animated characters are seen flying over the Twin Towers.
It’s a fact: we now know that internet users know all the tricks. DALL-E 3 is integrated into Bing Chat for free, and many have already started to “play” with its capabilities. However, the new censorship rules imposed by OpenAI have led some to see it as a challenge: what are the real limits of Bing Chat? Is it truly capable of avoiding offensive generations? It seems not.
A very striking example is that of the famous Twin Towers. The active blocking of words like “Twin Towers” or “World Trade Center” has not prevented the creation of scenes that include SpongeBob crashing into the buildings, to mention just one of many.
These measures have done very little for Microsoft. Caitlin Roulston, Microsoft’s Director of Communications, told The Verge that the company plans to improve its systems to “help prevent the creation of hurtful or sensitive content […] as with any new technology, some want to use it in ways that weren’t intended. That’s why we’re implementing a series of filters to make Bing Image Creator a positive experience.”
Microsoft has not provided details on what kind of filters would be used or how they would work. Moreover, these “blocks” are more urgent than ever. Users on forums like 4chan are already posting detailed guides on how to bypass censorship in Bing Chat and Stable Diffusion to distribute racist images, for example.
In the case of OpenAI, they themselves admitted that their security measures were far from “perfect” and are constantly being updated. Will we see the same thing soon with Bing Chat? Will their detection tools improve in the near future?
