Elon Musk’s xAI announces the open-source release of its AI chatbot Grok, available on GitHub as of 11 March. This move allows researchers and developers to contribute to and enhance Grok, impacting future updates as it competes with counterparts from OpenAI, Meta, Google, and others in the tech industry.
The open release comprises the ‘base model weights and network architecture’ of Grok-1, a Mixture-of-Experts model with 314 billion parameters. Notably, this model is from a checkpoint in October and has yet to undergo fine-tuning for specific applications like dialogue, as outlined in a company blog post.
Under the Apache 2.0 license, the release permits commercial use but excludes the training data and real-time data connections to xAI. The company originally introduced Grok in November 2023, positioning it for coding generation, creative writing, and question-answering purposes.
Elon Musk’s stance on open-sourcing AI models follows his acquisition of X (formerlyTwitter), where he advocated for transparency in algorithmic operations. Musk has criticised companies, including OpenAI, which he co-founded, for not adhering to open-source principles. This criticism culminated in a lawsuit against OpenAI for allegedly breaching their funding agreement.
While some companies release open-source or partially open-source models to solicit feedback for improvement, most widely used AI foundation models remain closed-source or offer limited open licensing. Meta’s Llama 2, for instance, provides free research but imposes commercial use fees, restricting developers from building upon it.
Grok’s initial launch required an X subscription to offer a more contemporary and engaging chatbot alternative to options like OpenAI’s ChatGPT and Google’s Gemini. However, early testing indicated it needed more sophistication and distinctiveness to compete effectively in the chatbot landscape.