Working as a designer for a LLM AI-powered product, which we'll refer to as 'AI' in this post, I have had the opportunity to try and use various AI products available in the market; for example, ChatGPT, Pi, Perplexity, Google Workspace, Amazon, Shop, TLDR, etc. My role involves creating a chatbot interface, and through my experiences, here are seven observations about AI-powered chatbots.
-
AI is always responsive
When unable to generate a sufficient answer, AI acknowledges its understanding of the query and indicates its inability to respond, which is deemed an acceptable form of communication in a conversation. Unlike traditional customer support chatbots that often triage users, AI stands out by never leaving me hanging. It engages me in open-ended conversations, readily responding to a wide range of inputs. Whether my queries are vague, out of context, or complex, AI consistently keeps the conversation going. -
AI understands imperfect inputs
AI comprehends and corrects imperfect inputs, including typos, incomplete grammar, and vague expressions. This capability distinguishes it from traditional keyword searches on websites, which are less forgiving of typos. Additionally, it surpasses conventional search engines; with AI, I can simply ask "What's that?" in a chat thread, and it accurately understands the context and provides relevant information. AI also asks clarification questions when needed. -
AI takes terse and natural language
AI seamlessly accommodates a wide range of input styles, from terse, single-word commands to complex, paragraph-length queries. Whether I request troubleshooting with just a word or provide a detailed situation description, AI responds with its understanding and relevant answers, ensuring a smooth and continuous conversation. -
AI covers a wide range of topics
The AI models powering AI products are typically trained on extensive knowledge bases, often sourced from the web. Even though a product incorporating these models may not have been trained specifically on technical terms featured in the product, its AI is often capable of providing explanations for these terms right out of the box. It's important to note that AI can not only answer questions about knowledge specific to the product but also on a broad spectrum of subjects. This includes everything from recipes and coding to life advice, which might sometimes seem contextually unrelated, depending on the product. -
AI is very useful at summarizing and identifying patterns
AI is good at summarizing content, identifying patterns, and providing valuable insights. When faced with lengthy or complex responses, I can request rephrasing and clarification as needed. Additionally, it can generate simple formatting, like tables, which enhances understanding, especially for complex information. By the way, the formatting could benefit from additional visual enhancements. -
AI goes non-sensical
AI, at times, provides fabricated or inaccurate responses. This can be considered both a limitation and a feature (in a use of creative activities). When developing a product, maintaining continuous QA is crucial, and having an ever-evolving checklist is essential. -
AI is overly accommodating and agreeing
AI's accommodating nature can sometimes lead to inconclusive conversations, necessitating mechanisms for meaningful conclusions.
All these features, except for #6 and #7, foster continuous conversation, delve deeper into topics, and maintain its flow. When developing a product intended to provide answers and assist users in achieving their goals, the AI in that context should proactively guide the conversation toward resolution. How might we ensure that each response is genuinely useful and easily digestible? How might we consistently propel the conversation forward and conclude it to a satisfying outcome? And, what strikes you in your world?