Billionaire Elon Musk-led artificial intelligence startup xAI has announced the launch of Grok, a unique AI language model designed to answer a wide range of questions with a touch of humor. With the view to compete with OpenAI’s ChatGPT and other AI chatbots, Grok boasts real-time knowledge via the social media platform ????, formerly known as Twitter.
Grok, now a part of X, is currently in its “very early” beta phase, and rapid improvements can be expected in the coming weeks. xAI is offering early access to a limited number of users in the United States to gather their feedback and improve Grok’s capabilities further. Users can join the Grok waitlist and be part of shaping the future of this exciting AI tool.
Talking about its journey, the company says that the engine powering Grok is Grok-1, a frontier Language Model (LLM) developed over the last four months. Grok-1 builds on the previous Grok-0 model, which already demonstrated remarkable capabilities with 33 billion parameters. In recent months, xAI has made substantial improvements in reasoning and coding capabilities, achieving impressive scores on various machine learning benchmarks such as GSM8k, MMLU, HumanEval, and MATH.
“On these benchmarks, Grok-1 displayed strong results, surpassing all other models in its compute class, including ChatGPT-3.5 and Inflection-1. It is only surpassed by models that were trained with a significantly larger amount of training data and compute resources like GPT-4. This showcases the rapid progress we are making at xAI in training LLMs with exceptional efficiency,” the company said in a blog.
— Elon Musk (@elonmusk) November 5, 2023
The mission of xAI is to create AI tools that benefit humanity as a whole. Grok represents a step toward that goal by providing a unique, user-friendly AI experience that aims to be inclusive of all backgrounds and viewpoints. It’s designed to be an indispensable research assistant, helping users quickly access information, process data, and generate new ideas.
xAI’s commitment to infrastructure reliability is evident in the development of Grok. They’ve built a custom training and inference stack using technologies like Kubernetes, Rust, and JAX. This infrastructure ensures that training runs smoothly, even at the scale of tens of thousands of GPUs, and effectively handles various failure modes.
Furthermore, the company says it is dedicated to addressing the limitations of current AI systems by advancing research in several key areas. They are exploring scalable oversight with tool assistance, integrating formal verification for safety, enabling long-context understanding and retrieval, enhancing adversarial robustness, and expanding Grok’s capabilities to include vision and audio. These research directions aim to make AI more reliable, intelligent, and capable of contributing to society in a meaningful way.