Generative AI (GenAI)

Modern generative models are increasingly "multimodal," meaning they can process and generate across different types of data simultaneously. For instance, a model can take a text prompt and output an image, or take an image and answer text questions about it.

• • Text-to-text: Drafting, summarizing, and translation (e.g., LLMs).
• • Text-to-image/video: Generating media from descriptions (e.g., Midjourney, Sora).
• • Text-to-code: Assisting software engineers with code generation (e.g., GitHub Copilot).

Organizations adopt GenAI primarily for operational efficiency, creative augmentation, and personalized customer interactions. Successful deployments often start with internal "copilot" use cases before moving to customer-facing applications due to safety considerations.

Generative models predict patterns; they do not possess true understanding or factual recall. This leads to hallucinations. They also lack the ability to verify the intellectual property of the data they generate.