If Musk wants AI for the world, why not open-source all the Grok models?

Elon Musk's xAI, the startup that owns the X (formerly Twitter) social media platform and builds AI models, unveiled its latest innovation, Grok 3, on Monday. This AI model challenges the best from OpenAI in reasoning and other tasks. 

The unveiling follows Musk's hostile bid for OpenAI, Inc., the non-profit that controls OpenAI, last week. Musk released a statement about the bid, saying "It's time for OpenAI to return to the open-source, safety-focused force for good it once was." 

The latest move is part of a campaign Musk has waged against OpenAI and its CEO, Sam Altman, Musk's former partner in the company's founding. Musk has repeatedly accused OpenAI and Altman -- including in lawsuits -- of abandoning the lofty founding vision of maintaining AI for the public good.

Also: xAI's Grok 3 is better than expected. How to try it for free (before you subscribe)

The OpenAI board declined Musk's offer.

This raises the question: If Elon Musk is concerned about keeping AI development transparent and safe, why hasn't he committed xAI to open-source Grok for all of its models starting right now, in perpetuity?

To date, xAI has only posted one open-source version of Grok, the first version, Grok 1, which is available in the xAI repositories on GitHub and on HuggingFace. Since that time, the world has seen Grok 1.5, Grok 2, and now Grok 3. None of these other models have so far been released as open-source models. 

Researchers and programmers debate the term "open-source AI models." The historical meaning of "open source" has indicated the release of a program's source code. 

In the case of open-source models, including Grok 1 and Meta's Llama AI models, no source code is released. Instead, the firms release only the model "parameters," or "weights," for a model. 

Because a lot can be re-constructed from having access to the weights, the AI field has been willing to bend the definition of open-source. Just know that this usage is not without controversy.

At press time, I had not received a reply to my request for comment from xAI or to my tweet on X directed at Musk, asking why he wouldn't commit to open-sourcing all xAI models.

Also: Elon Musk's X now trains Grok on your data by default - here's how to opt out

The first model, Grok 1, was released in November of 2023 and made open-source the following March, a four-month lag. The company quickly followed with Grok 1.5 the same month, then Grok 2 this past August, and now Grok 3. That suggests there has been ample time to make Grok 1.5 and Grok 2 open-source if the company were to follow the same cadence. 

Of course, companies can maintain a balance between making available open-source models and keeping other models closed. It's possible to both give back to the community and the world and also retain valuable intellectual property behind closed doors.

Also: All X users can access Grok AI chatbot and its new image generator now - for free

Musk's lack of a firm commitment or even a vague statement of principle is odd, given his professed concern for transparency in AI for humanity's sake.

By making a hostile bid for OpenAI, Musk seems to have shifted the burden of protecting humanity outside of his own corporate interests. Perhaps not surprising, since xAI has its own for-profit investors to please.

While Altman works to raise perhaps $44 billion for OpenAI's massive compute needs, xAI announced in December that it had raised $6 billion from venture capitalists in a deal that valued the company at $45 billion.  

Also: These nations are banning DeepSeek AI - here's why

With the sudden surge of interest in China's DeepSeek model last month, the handwriting is probably on the wall for all closed-source models, including OpenAI's GPT family and Musk's Grok. Ultimately, more and more AI innovation, at least of the large language variety, will become open-source, much the way the Linux movement has come to dominate operating systems, programming frameworks, and more.