The supposed ‘open sourcing’ of advanced artificial intelligence (AI) programs, such as Grok 2.5, has been called into question due to restrictive licensing. Such claims of open sourcing, recently touted by tech visionary Elon Musk among others, have been criticised as merely an instance of ‘open-washing’ – falsely advertising a product as open source when it doesn’t truly meet the criteria. While open source projects represent a goldmine of possibility, ensuring more equitable access and collaborative development, these half-hearted efforts by tech companies to appear generous are a concerning trend.
In a world where being perceived as ‘open source’ can dramatically increase company standing, particularly among developers and investors, the truth is often distorted. We’ve seen companies ride the open source wave before: first with the dubious claims that Meta Llama was made open source, and now with reports that popular AI startup xAI has open sourced their Grok 2.5 model, a large and sophisticated language model from last year.
“The xAI Grok 2.5 model, touted as our most impressive model from last year, is now open source. Grok 3 is set to follow suit in roughly six months’ time”, these were the words floating around during the announcement. The company released not just the model itself, but also all of its associated weights. One might question the motivation behind such a conspicuous move.
Unofficially, whispers abound that the move is aimed at luring more people towards Grok, antagonistically pitting it against its competitors. Critics identify this as classic open-washing: an operose attempt to ride the open source hype train without genuinely liberating the code in the true open source spirit. On the record, however, xAI contends that this is an attempt at promoting transparency and inspiring widespread developer participation in their coding processes.
If you’re wondering, ‘What’s the catch?’ – the answer lies in the Grok license. In theory if you enhance the code, xAI would gladly adopt your modifications. That seems normal for any open source project, but the fine print creates a different picture: “You may not use the Materials, derivatives, or outputs (including generated data) to train, create, or improve any foundational, large language, or general-purpose AI models, except for modifications or fine-tuning of Grok 2 permitted under and in accordance with the terms of this Agreement.” So much for being open source, eh?
As one observer astutely pointed out, this essentially means, the source isn’t really available, so it’s not truly ‘open source’. There are restrictions on the model weights’ usage, so they can’t be characterised as ‘open weights’. What we have here is a situation where model weights are available, end of story. The open source label doesn’t firmly apply here.
Indeed, the Grok model fails to meet the criteria of the Open Source Initiative (OSI) Open Source AI Definition (OSAID). This extends further still, as it also doesn’t fulfil the requirements of other widely recognised open source definitions. As it has been resoundingly echoed in tech circuits, any license that places restrictions on the field of use falls short of the open source definition.
Perfect examples of models that yield to the Open Source Definition and OSAID are the Olmo2 and Molmo models developed by the Allen Institute for AI, along with others listed in the OSAID FAQs. They are a beacon of true open-source projects in contrast to the likes of Grok.
The restrictions in Grok’s license make it unworthy of the open-source label. First, it conditions commercial use on ‘guardrails’, meaning the user has to comply with every point in xAI’s Acceptable Use Policy. This is a step beyond regular OSI-approved licenses. Second, it prohibits using the materials for training, creating, or improving any foundational, large language, or AI models barring limited modifications to Grok 2 itself. This discriminates against certain fields of endeavour, contradicting OSI’s criteria.
Further adding to its list of cons is the so-called ‘termination clause’, where the license termination is triggered if certain types of litigation are pursued by the licensee. This clause is far more restrictive than any OSI-compliant license would suffice.
Given these restrictions, one naturally wonders: what’s Grok good for? Well, you’re free to run, study, and modify Grok 2.5. xAI asserts that this encourages independent exploration, potential advancements, and transparency in how complex AI systems are constructed. It’s not as dismal as it first seems, but it’s certainly not completely open.
There exists an abundance of genuinely open AI projects such as Mistral, Phi-2, BLOOM, and GPT-OSS. These provide unrestricted access, allowing you to get a true understanding of how AI systems function. They provide a far greater level of openness than what Grok has to offer.
In summary, if Grok piques your interest, dive right in. Explore its workings, attempt to enhance it, and learn from it. Enjoy the process, but tread carefully. It is important, however, to not delude oneself – despite what the marketing might suggest, you’re not dealing with a fully open source code or open weights. Actual open source projects have a different ethos. And it is important for us to recognize the difference.
The post Advanced AI’s Deceptive Open Source Claims appeared first on Real News Now.
About Author
