Open vs. Closed AI Models: What Businesses Need to Know

Choosing the right Artificial Intelligence (AI) model is a big decision for businesses. It's not just about the technology; it's about the overall strategy. Companies need to think about whether to use open, closed, or a mix of both types of AI models. Experts from companies like General Motors (GM), Zoom, and IBM recently discussed this at an event called VB Transform, sharing how they and their customers make these choices.

According to Barak Turovsky, GM's Chief AI Officer, there's a lot of hype around new AI models. He pointed out that making AI model information openly available (open-sourcing) has led to major advancements. He humorously noted that open-source principles initially helped companies like OpenAI develop, which then moved towards more closed approaches, and may now be trending back towards openness.

Several things influence the decision, including how much it costs, how well it performs, and how trustworthy and safe it is. Turovsky mentioned that many companies prefer a mixed approach. They might use open models for internal tasks but switch to closed models for customer-facing applications, or vice versa.

Armand Ruiz, VP of AI Platform at IBM, explained that IBM started with its own AI models. However, they realized they needed more options, especially as more powerful models became available. So, they added support for platforms like Hugging Face, allowing customers to choose any open-source model. IBM even launched a tool that lets companies easily switch between different AI models.

More and more companies are buying AI models from multiple providers. A survey of CIOs showed that a significant percentage are using five or more different models. While having choices is good, too many choices can be confusing, according to Ruiz. IBM helps customers by focusing on the specific problem they're trying to solve first, without getting too caught up in which AI model to use initially. They focus on whether the project is feasible. Later, they decide whether to fine-tune a model or customize it to meet the customer's specific needs. In essence, IBM helps businesses cut through the complexity and find the best solution for their needs.

Zoom offers its customers choices for its AI Companion. Customers can choose to combine Zoom's AI model with other, larger models. Or, if they're concerned about using too many models, they can stick with just Zoom's model. Zoom has also partnered with Google Cloud to improve AI Companion for business tasks.

Zoom created its own small language model (SLM) without using customer data. This model is surprisingly small but can still outperform other models designed for specific industries. It works best when used with a larger model for complex tasks.

Zoom CTO Xuedong Huang emphasized the power of this mixed approach. He likened it to "Mickey Mouse and the elephant dancing together," where the small model handles specific tasks while working as a team with the larger model. This illustrates the idea that a smaller model isn't necessarily sufficient on its own, but it can be very effective when combined strategically with a larger model.