The integration of artificial intelligence (AI) into business processes is witnessing a significant transformation, with both open source and closed source models playing pivotal roles in this evolution. The discussion is increasingly emphasising the distinctions between these models, their implications for businesses, and the associated challenges of sustainability, customisation, and legal liabilities.
Sreekanth Menon, global head of AI at Genpact, suggests that the future landscape is likely to be hybrid. He indicated that despite the prevalent belief in an open-source dominance, "both open and closed-source models have their place". He further noted that enterprises would benefit from being "model agnostic", allowing them to leverage the strengths of various models without being confined to one approach. Closed source models are typically backed by substantial investment, enabling them to deliver specialised solutions that are informed by deep research and development.
The discussion around open source AI has intensified recently, particularly with references to Meta's Llama models. Although these models are widely considered to be open, they do not fully conform to the newly established definition of open source AI by the Open Source Initiative, which was released in late October. According to this definition, open source AI must provide comprehensive sharing not only of source code but also of model parameters, training data characteristics, and usage rights without needing prior permission.
Mark Collier, COO at the OpenInfra Foundation, explained the importance of comprehending these distinctions, stating, "To me, what matters most is that people and companies have the ability and freedom to take this fundamental technology and remix it, use it, and modify it for different purposes without having to ask a gatekeeper to give them permission." He warned procurement teams to scrutinize vendor claims about open source technologies, as there can often be overlooked restrictions, such as those associated with the Llama models, which require special licensing for broader usage.
The complexity of customising open source models is underlined by experts in the field. Raghavan Chandrasekaran noted to CIO that open source models require significant investment in retraining and modification, with rapid changes in base models creating additional complications. "If you customize something and the base model changes, you have to re-customize it," he added. This highlights the ongoing challenge businesses face in ensuring their AI systems remain relevant and effective over time.
Furthermore, the sustainability of open source AI models remains uncertain. While initial versions may gain popularity, the long-term financial support for ongoing development is less certain. Chandrasekaran pointed out the difficulties in monetising such models, raising the question of who will continue to provide funding for future iterations.
Amidst this evolving landscape, corporate users are urged to be vigilant about licensing and the implications of customisation, particularly concerning the confidentiality of proprietary data. As licensing practices develop and change swiftly, it is crucial for companies to align their strategic decisions with a clear understanding of both open and closed source models.
Source: Noah Wire Services