Grok-1, Elon Musk’s artificial intelligence model, is now publicly available under the Apache 2.0 license thanks to an official open-source initiative by xAI. Following Musk’s declaration that he intended to make Grok open-source, this action aims to democratize access to cutting-edge artificial intelligence technologies. xAI was used to train the 314 billion parameter Mixture-of-Experts model known as Grok-1.
The network architecture and basic model weights are included in the open-source release; however, the fine-tuning code and training datasets are not. A magnet link for a torrent file containing 773 files and roughly 300GB in size is provided to download the model. This statement is a part of Musk’s larger attack on AI firms, such as OpenAI, which he co-founded but eventually left, for straying from open-source ideals.
Powered by Grok-1, the Grok chatbot was first made available to paying users on X (previously Twitter) in November. Its goal is to respond to queries with a sense of humor and a rebellious demeanor. It competes with other AI chatbots by providing real-time information and a distinct sense of comedy, such as OpenAI’s ChatGPT. Grok-1 has shown competitive performance on a number of benchmarks, with a score of 62.9% on the GSM8k benchmark.
Unlike other AI models that offer more limited access, xAI hopes to enable wide access to its underlying technology by making Grok-1 open-sourced. This announcement is noteworthy in light of the growing hostilities between Musk and OpenAI and his continuous condemnation of AI firms for putting profits ahead of ethics and openness.