Generative AI innovation typically starts with choosing a pre-trained ‘foundation model’ and then customizing it to excel at a specific task. However, the range of use cases is so vast, and generative AI technologies are maturing so rapidly, that it can be hard to match use case to model – and to keep up with the latest and greatest.
Foundation models are highly differentiated from one another. Some have use case specializations (chat, code, analysis, summarization), others have modal specialization (text, image, speech, multi-modal), still others have operational specializations (intelligence, cost, latency, etc.).
Some foundation models will outperform others on specific tasks or in terms of output quality. Meanwhile, chaining models together into an agent or intelligent system allows you to lean into the specialization and uniqueness of the individual models while driving up the aggregate usefulness of the system.
