The new generative AI model from xAI, Grok 3, is approaching its official release. Several users of the X platform reported that they had gained access to Grok 3, the successor to Grok 2, through the Grok app. Although access was quickly blocked, users managed to test the model on logical reasoning and programming tasks.
Grok 3 is capable of successfully solving puzzles and handling code generation requests, such as for a casino roulette. However, some flaws and errors were found in the roulette code. The model also impresses with its ability to answer complex questions that other models struggle with, although it sometimes demonstrates overconfidence.
The Grok 3 model has been in development for several months, and Elon Musk noted that it has completed the pretraining stage. Grok 3 is expected to be officially released in January or early February. A large data center in Memphis with approximately 100,000 graphics processors is being used for its training.
Grok 3 potentially features new capabilities, such as a “voice mode” that allows the model to vocalize its responses. It can also analyze images and answer questions, making it a competitor to models like GPT-4 from OpenAI and Gemini from Google.
