On Tuesday, Elon Musk’s AI company, xAI, introduced the beta release of two new language models, Grok-2 and Grok-2 mini, for subscribers of his social media platform, X (formerly Twitter). These models are integrated with the recently launched Flux image-synthesis model, enabling X users to create largely unrestricted photorealistic images that can be shared on the platform.
Ethan Mollick, a frequent AI commentator, highlighted the potential issues with Flux, noting that it excels at generating realistic images, including fake photos of actual locations and people. “Does anyone know if they are watermarking these in any way? It would be a good idea,” Mollick questioned on X.
A report from The Verge indicated that Grok’s image-generation capabilities come with minimal restrictions, allowing users to create potentially controversial content. During their testing, Grok produced images of political figures in compromising scenarios, copyrighted characters, and violent scenes. Despite claims of limitations against certain content, the rules appear inconsistently applied.
The Verge’s findings show that, unlike other AI image generators, Grok does not seem to reject prompts involving real people or add identifying watermarks to its outputs. This has led to the creation of images such as Donald Trump and Kamala Harris in fabricated situations, raising concerns over the lack of oversight and the potential legal implications.
Musk, in response to questions about why xAI chose to use Flux without additional content filters, stated, “We have our own image generation system under development, but it’s a few months away, so this seemed like a good intermediate step for people to have some fun.”
Table of Contents
Grok-2: Expanding the GPT-4 Class, But No Breakthrough Yet
Beyond image generation, xAI claims that Grok-2 and Grok-2 mini represent significant advancements, with Grok-2 reportedly outperforming some leading competitors in recent benchmarks. Despite these claims, the “GPT-4 class” of AI models remains unbroken, with Grok-2 joining a lineup that includes GPT-4o, Claude 3.5, Gemini 1.5, and Llama 3.1.
According to xAI, Grok-2 shows improvements in areas like graduate-level science knowledge, general knowledge, and math problem-solving. It also reportedly excels in visual tasks, achieving top results in visual math reasoning and document-based question answering.
Both Grok models are available to X Premium and Premium+ subscribers through an updated app interface, but unlike some competitors, xAI is not releasing the model weights for independent verification. The company plans to offer both models through an enterprise API later this month, with security features like multifactor authentication, though details on pricing and data-handling policies remain unclear.
