xAI releases Grok-2 with controversial image generation
Shortly after a release, there are again discussions about the Grok image AI. This time, they concern the integrated AI image generator.
Grok's picture testimonies are sometimes without guarantee.
(Image: Erstellt mit Flux/Grok-2 durch Alejandra Caraballo)
Elon Musk's AI company xAI has unveiled a new version of its chatbot Grok. Grok-2 promises enhanced capabilities in the areas of chat, programming and logical thinking. In its announcement, xAI promises better results than GPT-4 Turbo and Claude 3.5 Sonnet, for example. There is now also an image generation function, which was developed together with the German start-up Black Forest Labs. "In collaboration with Black Forest Labs, we are experimenting with their Flux.1 model to extend the capabilities of Grok to X," reads the xAI announcement.
Controversy over uncensored content
The images generated by Grok-2/Flux.1 caused controversy after just a few hours, as there are hardly any restrictions in contrast to the competing products from OpenAI or Google. Users could easily create controversial and potentially misleading images, such as a marijuana-smoking Donald Trump, Ronald McDonald with a machine gun in front of Burger King, burning McDonald's branches and other more or less tasteless material.
Musk had also decided that his chatbot hardly needed any restrictions. The chatbot itself claims not to generate any pornographic, excessively violent or copyrighted content. In practice, however, there appear to be hardly any safety precautions.
Technology from Black Forest Labs
The image generator from Black Forest Labs has been attracting attention for several weeks with its particularly realistic and detailed results. According to the Artificial Analysis platform, the Flux.1 model outperforms the image generators Midjourney v6.0, Dall-E 3 from OpenAI and SD3-Ultra from Stable Diffusion in some benchmarks in terms of the quality rated by users. This includes visual quality, prompt compliance, typography and output variety. The founders of Black Forest Labs were previously involved in the development of Stability AI's Stable Diffusion models. The recently founded start-up received 31 million dollars in seed capital. In addition to image generation, Black Forest Labs is also working on a video generator.
If you want to test Flux.1 in Grok-2, you need a paid premium account. Flux.1 can also be tested independently of Grok-2, as Black Forest Labs also offers its own models as open source.
Regulatory challenges
Grok-2's lack of security mechanisms could lead to regulatory problems for xAI and Platform X (formerly Twitter). In the US, legislators are looking for ways to regulate AI-generated counterfeiting and disinformation. The European Commission is already investigating X for possible violations of the Digital Services Act.
Videos by heise
So far, X has not commented on the allegations. However, since the beginning of August, xAI has been looking for reinforcements to monitor and evaluate the quality problems of extensive data collections. Following controversy surrounding the results of Gemini's image generators – there were "glaring historical inaccuracies", Google temporarily paused the service at the beginning of the year.
The European Commission is already investigating X for possible violations of the Digital Services Act. In the UK, the regulatory authority Ofcom is also preparing to enforce the Online Safety Act, which could also affect AI-generated content.
(mack)
