Without any questions, the Twitter chatbot generated a naked Taylor Swift, Donald Trump in a Nazi uniform, or Barack Obama with a cocaine track.
Currently, Grok, the chatbot that generates the images, is only available to X Premium subscribers, but it seems to operate without any moral or ethical constraints — despite the fact that it claims that «avoids pornographic images, dipshits, or media that violates»’s copyright.
However, based on the images that have been shared on Twitter and tests conducted with the generator by The Verge, these restrictions can sometimes be circumvented. Among the requests that were successful:
Of course, Grok — is not the only option for getting «banned images». There are still open source tools available, such as Stable Diffusion, which can be customized to do anything with minimal restrictions. Other companies tend to take errors quite seriously: Google, for example, had to suspend Gemini’s operation after issuing images with racial and gender stereotypes.
However, given Elon Musk’s love of «freedom of speech», it is unlikely that Grok will receive any corrections in the near future, unless the social network is threatened by EU or US regulators. The European Commission, for example, is already investigating Twitter for violation of the moderation rules and earlier this year requested information from the company on efforts to mitigate risks associated with artificial intelligence.
Source: The Verge, Ars Technica