Grok’s new text-to-image tool has few guardrails, as users are posting everything from cartoon characters holding assault rifles to US presidential candidates committing acts of terrorism.
xAI’s chatbot Grok has gotten image generation capabilities and the result is as messy as one would expect from a “rebellious chatbot”.
The new text-to-image generator bears similarities to many of the popular examples witnessed in recent years such as OpenAI’s Dall-E models. Users can type in a description of the image, add in specific details or styles they want and Grok will create an image thanks to a massive amount of training data.
Where Grok clearly differs is in terms of what it will allow – the initial results being posted on X show this image generator has very few guardrails when it comes to creating offensive imagery, using copyrighted content or creating images of public figures in bad scenarios.
Many AI-powered image generators have various guardrails to prevent this type of content – a user will struggle to create an image of a specific Disney character from Dall-E for example. With Grok, images have been shared of copyrighted characters holding assault rifles or committing murder.
AI-generated images are also being shared of US presidential candidates Donald Trump and Kamala Harris, in various offensive scenarios.
More normal image examples show the capabilities of the image generator – asking Grok to create images of people can create very accurate results that would rival other major image generators.
But the lack of guardrails could cause serious problems for xAI and its owner Elon Musk – companies won’t like seeing their copyrighted characters in controversial scenarios and neither will public figures. Musk himself may not like the results, as various have created AI images of the billionaire in embarrassing situations.
Some users have shared issues with the Grok AI model and claim that its limited guardrails can be easily bypassed to create even more offensive imagery or get instructions for illegal activities.
The Grok chatbot told The Verge that it has guardrails, but it appears to be a generic paste of what other chatbots would say, as it does not appear to follow any of the rules it listed in its reply.
Grok access continues to be limited to Premium and Premium+ users on X. The wave of AI-generated images comes while X is being scrutinised for how it has handled the spread of misinformation on the platform and its plans to collect user data to train its AI models.
Find out how emerging tech trends are transforming tomorrow with our new podcast, Future Human: The Series. Listen now on Spotify, on Apple or wherever you get your podcasts.
