The adoption of generative AI, whether for text, images, videos, or a combination of these, is no longer a matter of "if" but "when" for organizations. The interest in generative AI has surged in recent times, driven by its potential to revolutionize various aspects of business operations. However, as organizations explore this transformative technology, they face the challenging decision of whether to build their generative AI systems from the ground up or buy pre-existing solutions. This choice is complicated by the myriad options available for refining and customizing purchased services and the significant effort required to turn a bought or built system into a reliable and responsible part of the organization's workflow.
The Three Approaches: Taking, Shaping, and Making
The Taker Approach: In the "taking" model, organizations consume generative AI through APIs or other applications, such as ChatGPT or GitHub Copilot. This approach is suitable when adopting generative AI features and capabilities in existing applications is crucial for maintaining competitiveness. These readily available features can help organizations accelerate their workflow and provide new functionality, but they may not offer extensive customization.
The Shaper Approach: Shaping generative AI models involves leveraging existing foundational models available off the shelf. However, these models are retrained using an organization's own data to enhance accuracy and relevance. Shaping is particularly effective in reducing the "hallucination" problem commonly associated with generative AI. By recalibrating models to their specific needs, organizations can improve the quality of results.
