Leveraging LLMs in product design: opportunities and challenges
Incorporating AI into your design solutions.
Recently, I was making a lot of AI prototypes, and in this article, I want to share with you my learning. We are going to discuss what design opportunities Large Language Models (LLMs) unlock and how to design with AI.
Chatbots
LLMs have certainly increased the possibilities of chatbots. They can be handy when users just want the answers to their questions. For example, a chatbot can recommend products and provide additional relevant information to users.
But chatbots are not always the best type of interaction:
- It’s passive, users have to initiate the interaction
- Users don’t always know what they can ask a chatbot
- Users need to do prompt engineering
- The answers might not be useful for the users
Let’s see what other types of interaction are possible with LLMs.
Text-to-Text
LLMs take text as an input and provide text as an output. This can be creatively implemented in numerous ways.
Enhancing User Experience with Shortcuts
Prompt engineering is crucial for effective LLM interactions, but users cannot be expected to engineer their prompts. Consider how users are supposed to understand what they can do with your AI solution.
Identifying the Right Moment of Interaction
Identifying when a user needs help and providing timely assistance can drastically improve the user experience.
Instead of making one knowing-all-chatbot consider an AI assistant that can be invoked in a particular moment and perform narrow tasks that help users.
For example, Robin AI uses LLM to evaluate particular parts of a contract and to suggest new, alternative language that’s more friendly to our customers. The suggestions are shown when users select the text in the contract.
Using Context
Large language models can use context, which means they can utilize your data to provide a better user experience.
For example, Google plans to use LLM with the documents from the Drive as the context information.
When the context is used, your data is stored in your space. LLM is able to find relevant chunks of information in a vector store based on user request.
For more details, read my article on technical insights and limitations:
Enhancing Search Capabilities
Large language models can enhance search capabilities, making it more valuable for users. They can provide follow-up questions to refine searches or offer useful suggestions alongside search results.
Personalization with AI
With AI, personalization can be taken a step further. For instance, a search for a summer dress can be transformed into a personalized, editorial-quality page based on user input.
Mapping Natural Language to Other Domains
LLMs can understand only text for now, but this text can take various forms, broadening the potential for language mapping to other domains
The inherent flexibility of LLMs stems from their ability to process diverse forms of text input. This can range from a straightforward user query or a piece of code to context information like the user’s current position on a webpage.
Mood to Smart Home Commands
Imagine a user interacting with a smart home assistant enhanced by an LLM. After the user expresses a desire to relax, the system suggests suitable adjustments to create a relaxing ambiance, such as dimming the lights and playing a calming playlist.
The system is able to do so because it knows available smart home commands, has access to playlists, and is able to map user requests to available commands.
Natural Language to UI
Galileo AI maps natural language to the design layouts in Figma.
I don’t work at Galileo, but the possible implementation might be as follows. The LLM assesses the user’s requirements, selects appropriate design elements, and assembles a layout. The design elements might be as small as buttons, avatars, and cards or as big as pre-design mobile screens.
Recap
Understanding the potential of LLMs can stimulate innovative solutions to everyday challenges. LLM is not the only available AI technology, there are many other AI types.
Solve for the use case, AI is just a tool
Any technology, including AI, serves as a tool to solve problems rather than an end in itself. The focus should always be on identifying the solution to a particular use case rather than seeking a problem that fits a specific technical capability.
When brainstorming, don’t be constrained by perceived technical limitations. It is often simpler to refine an ambitious concept than to expand a restricted one. In essence, technical knowledge should be used to widen the scope of possible solutions rather than narrow it down.
