Deciding on the right artificial intelligence (AI) models can be a complex dance between technical prowess and strategic vision. At this year’s VB Transform conference, experts from leading companies like General Motors (GM), Zoom, and IBM shared insights into the nuances of choosing between open, closed, or hybrid AI models.
Barak Turovsky, GM’s inaugural chief AI officer, highlighted the evolving landscape of AI models. He emphasized how each new model release and leaderboard shift introduces a wave of excitement and challenges. Turovsky reflected on his role in launching one of the earliest large language models (LLMs) and underscored the pivotal role of open-sourcing AI model weights and training data in driving innovation.
Turovsky remarked,
“Open-source actually helped create something that went closed and now maybe is back to being open.”
This dynamic interplay between openness and proprietary technology underscores the fluid nature of AI model development.
The factors influencing decisions around AI models are multifaceted, ranging from cost considerations to performance metrics and issues related to trust and safety. Enterprises often adopt a blended strategy where they leverage open models for internal processes while relying on closed models for customer-facing applications or vice versa.
Armand Ruiz, IBM’s VP of AI platform, shed light on IBM’s strategic evolution in response to the rapid advancements in AI modeling. Initially focusing on developing proprietary LLMs, IBM recognized the need to embrace a more diverse ecosystem as more powerful models entered the market. The company pivoted towards integrating platforms like Hugging Face to offer customers access to a broader array of open-source models.
Ruiz highlighted the growing trend among enterprises to procure multiple models from various vendors. This trend reflects a shift towards diversification in AI model selection as organizations seek tailored solutions that align with their specific needs.
Xuedong Huang, CTO of Zoom, delved into how Zoom approaches its AI Companion offerings for customers. The company provides two configurations – one that federates its own LLM with other foundational models and another option for customers preferring Zoom’s standalone model.
Huang emphasized the significance of hybrid approaches where smaller language models (SLMs) complement larger ones effectively. He likened this synergy between small and large models to
“Mickey Mouse dancing with an elephant,”
illustrating how each model size serves distinct purposes but works harmoniously together towards a common goal.
In navigating the labyrinthine world of AI model selection, Ruiz shared IBM’s customer-centric approach focused on simplifying decision-making during initial phases by prioritizing use case feasibility over intricate model comparisons. This approach aims at streamlining adoption processes by honing in on practical application scenarios before delving into customization or distillation considerations.
As organizations grapple with an expanding array of choices in the AI modeling arena, striking a balance between flexibility and clarity becomes paramount. While choice fosters innovation and tailored solutions, it also poses challenges around decision fatigue and complexity management.
Leave feedback about this