MeshThink

The rise of AI Large Language Models (LLMs) as a service (OpenAI, Gemini, etc) has led to tons of exploration with custom-built chatbots for ecommerce. Why haven’t we seen more AI Ecommerce Chatbots?

AI Large Language Models operate in way that forces tradeoffs between unrestricted Q&A and accuracy, and it’s some of these quirks that make these chatbot experiments unusable for production. ‘Hallucinations’ — making facts up — is a downside of using LLM approaches. And the way a question is asked can make a big difference in whether an LLM gives an answer that’s based in fact.

Unfortunately, no matter how well-trained you are in these nuances, a chatbot user interface pattern give a lot of weight to the user of the chatbot — someone who doesn’t know that how they ask a question matters! This makes it difficult — supporting any possible question with any possible phrasing without ever inducing hallucinations is a very difficult task!

We explore one of the most common triggers for this kind of behavior, and also talk about alternatives, exploring the use of LLMs in user interface patterns beyond simple chatbots.