Introduction:
The VentureBeat AI Impact tour recently stopped in San Francisco, where the focus shifted to the real-world applications of generative AI and the challenges of deploying gen AI initiatives. Keynote speakers included Ed Anuff, DataStax’s chief product officer, Nicole Kaufman, chief transformation officer at Genesis Health, and Tisson Matthew, CEO of Skypoint.
Moving Beyond the Experimental Phase:
Enterprises are moving past the experimental phase of generative AI and starting to explore how to integrate its power with their own business-critical data. According to Ed Anuff, an AI maturity model is emerging, indicating that companies are transitioning from one-off projects to critical AI business initiatives. These initiatives require careful deployment in high-impact, high-visibility areas, which may take more time, but have transformative potential.
Use Cases for Gen AI:
Generative AI has a wide range of use cases, from back-office operations to customer-facing applications. While terms like “chatbots” or “conversational interfaces” are still used, the underlying goal is to build knowledge apps that can retrieve information interactively. Organizations must decide whether to develop these apps in-house or use off-the-shelf products.
Considerations before Production:
For applications like customer support or financial analysis, off-the-shelf solutions can be effective. However, when addressing use cases critical to core business activities, customized data curation becomes crucial. Healthcare applications, for example, require real-time responses based on changing patient readings. In these cases, a custom AI application connected to core data assets is necessary, similar to how companies like HomeDepot or Best Buy build tailored websites.
Calculating Readiness and Cost:
As enterprises progress beyond the ideation stage, two primary issues arise: relevancy and cost. Relevancy refers to the appropriateness of AI responses and the retrieval of accurate content. Companies often struggle with these issues, leading them to reevaluate their data architecture. The cost of production is another important consideration, as surfacing relevant results can be expensive.
Hallucinations, Data, and the Importance of RAG:
“Hallucinations” occur when AI systems produce inaccurate or irrelevant responses. However, not all errors are hallucinations; some stem from training set errors. To mitigate this, the use of RAG (retrieval augmented generation) is crucial. RAG combines knowledge retrieval with generative AI, allowing for context-aware responses in natural language. By grounding responses in provided information, the chances of hallucinations significantly decrease. Additionally, RAG enables the integration of real-time company data accurately and securely into the AI model during inference.
Conclusion:
Deploying gen AI initiatives requires careful consideration of use cases, relevancy, and cost. Customized applications connected to core data assets are necessary for critical business activities, while off-the-shelf solutions can suffice for certain use cases. Mitigating hallucinations and ensuring accurate data integration are key challenges that can be addressed through RAG. As enterprises navigate the complexities of deploying generative AI, they can unlock its transformative potential in various industries and applications.
