xAI recently launched two new Grok models. This is the second version of the language model, and it teams with the social media platform X even harder.
Grok-2 is said to reason better and is capable of generating images. This compares to the 1.5 version of the chatbot released a few months ago. The model also comes in a second mini-version.
Grok-2’s capabilities were tested by xAI through Chatbot Arena. This platform allows anyone to ask questions of different LLMs and indicate their favorite answer. On this popularity test, Grok-2 does not match ChatGPT-4o.
Source: xAI
The model also passed a number of standard tests that tested mathematical understanding and reading comprehension, among others. Additionally, Grok-2 excels in vision-based tasks, delivering state-of-the-art performance in visual math reasoning (MathVista) and in document-based question answering (DocVQA),” is stated.
Political figures in graphics
In addition to claimed improvements in chat, coding, and reasoning, Grok-2 can generate images on X. The first results of that functionality are already being shared on the social media platform.
Users mainly notice that restrictions on image generation are not yet as strict as in other AI models. That makes it possible to create images of political figures.
It is unclear whether Grok-2 watermarks the fabricated images. OpenAI, for example, does this based on the C2PA standard, which puts metadata in the image.
Beta version, API to follow
Access to the LLM is only available through X. Users must also have a paid subscription with a Premium or Premium+ account. These users can access the two new models, although for now, this is a beta version.
xAI will release the API for both models later this month. Developers will find it through the new enterprise API platform.
Also read: Complaints filed in European countries against AI training Grok on X-posts