Deep learning can create meaningful representations of data, akin to "uncrumpling a paper ball" as described by Francois Chollet. These "embeddings" are stored in knowledge bases, enabling powerful applications. Imagine an insurance company creating embeddings of policy documents. These would represent the business rules, which are currently scattered across documents and applications. This new knowledge base would enable autonomous agents to handle tasks with greater efficiency and accuracy.
Furthermore, these embeddings could be used to eliminate biases, encourage creative thinking, and enable collaboration across the enterprise. Imagine relationship managers receiving personalized talk tracks for clients, artists creating movies with generative tools, and researchers using protein data to design new drugs. With deep learning, the possibilities are endless, and limitations such as hallucinations might become features, not bugs.