Companies are now shifting their focus from being the first to adopt AI to deploying these powerful tools effectively. This shift comes as firms realize that poorly crafted prompts and the use of unspecialized AI models can lead to inaccuracies and inefficiencies. To address this, companies like Johnson & Johnson are creating prompt libraries to improve the quality of AI output, while others like Starbucks are developing in-house models.
Creating effective prompts involves providing clear instructions and tweaking keywords to achieve desired results. Prompt libraries help minimize the risk of AI hallucinations and enable more efficient answer formatting. However, even with optimized prompts, there may still be limitations due to the absence of relevant knowledge in the data set.
To overcome these limitations, companies like Ernst & Young are taking additional steps. They have developed in-house AI platforms and leveraged specialized systems provided by partners like Microsoft to fine-tune AI models and adjust them to meet specific outcomes. Fine-tuning is done by machine learning experts, and companies may shrink datasets and create embedding libraries to further specialize the system.
By combining controlled datasets, embedding libraries, and customized prompts, companies can create hyper-specific AI models tailored to their purposes. This allows them to leverage collective knowledge across multiple jurisdictions and personalize AI models for enhanced performance.
While some believe that as AI becomes more intelligent, the importance of prompts will decrease, others argue that there will still be a need for specialized systems and engineered prompts for irregular tasks. The future is expected to involve a mix of specialized and generalized AI models, depending on the nature of the tasks at hand.