I was engaged in a conversation with a pre-seed startup leveraging AI over the weekend. With the current AI buzz, it’s become quite common in Bangalore for casual weekend brunches to morph into AI startup discussions. 😊

The discussion initially revolved around the usual strategy topics: Product-Market Fit, MVP products, Target Segmentation, and so forth, before shifting towards questions about AI.

One question that particularly piqued interest was whether they should implement RAG (Retrieval-Augmented Generation) or SLM (Small Language Model) to achieve the most efficient solution.

🚀 However, this isn’t an either-or discussion. It’s crucial to weigh both solutions, considering the specific use case for which the solution is being built.

✳ At first glance, RAG offers several advantages, such as providing accurate responses, better customization, and contextual awareness. However, one must also consider the costs involved — RAG implementation is more complex, computationally expensive, and requires maintenance.

✳ On the flip side, SLMs are straightforward solutions with a high degree of adaptability and the ability to provide creative yet novel responses. However, they lag due to their context limitations, occasionally providing incorrect answers and facing challenges around fine-tuning.

✅ Understanding your use case becomes crucial at this juncture. Does your solution require a high degree of accuracy and context, such as FAQ Chatbots, Technical Support, or Customized responses? If so, it’s worth investing additional resources and time in leveraging RAG to build a more robust solution.

Conversely, if your use case leans more towards creative and open-ended tasks, such as conversational chatbots or non-FAQ scenarios, opting for an SLM over RAG might be more suitable.

The founder (a good friend of mine) had some homework to do. It wasn’t the AI technique that was the focal point of the discussion, but rather the use case that drove the adoption of these techniques.

🔆 The power of AI lies in the techniques, models, and tools you adopt, providing desired results only when you spend time drafting your use cases accurately and comprehensively.