How LLMs are Transforming Conversational UX
- Chatbot Conference Team
- Sep 9, 2023
- 2 min read
Hello, Trailblazers of the AI Space,
Remember the days when building a bot felt like a never-ending cycle of setting flows and interpreting user inputs through rigid NLU frameworks?
Conversational Design Before LLMs
Not too long ago, we were all immersed in creating intricate flows, anticipating every possible user input, and crafting responses accordingly. It was a time of structure, precision, and rigidity—a time when creativity felt bound.
Those days are gone. They are becoming a distant memory, thanks to the introduction of Large Language Models (LLMs).
Conversational Design After LLMs
Today, the world of Conversational AI has been flipped upside down. The work is less about the exhaustive setting of flows and more about understanding the potent dynamics of prompt chaining and harnessing the power of well-structured knowledge bases.
However, the real change is in thinking.
Designers now need to think about what information is and how it is interpreted into something meaningful.
Understanding this dynamic will lead to organizing and retrieving information more accurately and efficiently and to creating better conversational agents.
So in summary, it boils down to:
Organing and Retrieving Information and generating Meaning
Flows: Very strategic Flows using NLP and NLU
The New Conversational Design Stack
The new CUX Design Stack will include a whole new section on Knowledge.
LLMs will tap into the Knowledge and be able to answer most questions, and then NLUs will follow up with the next logical step for the business.
For example, a user might ask about a product warranty, which triggers the Knowledge Base. The LLM answers the user's questions using the knowledge base and the user responds with enthusiasm. Then, the NLU is triggered to take the user down the happy path and close the sale.
In the example, the design stack includes:
Knowledge Base: Training the LLM on specific information
Prompt Design / Engineering / Tuning: prompts, instruction, roles given the LLM
NLU Flows: Very strategic Flows using NLP and NLU
Knowledge Bases and the Future of Conversational UX
To give you a visual guide through this significant shift in bot creation, we've prepared a video that highlights the nuances of using LLMs effectively.
Join us in exploring this evolving landscape, where we'll continually delve deeper into making the most of the technologies reshaping our industry.
I've spent years mapping intents and entities for voice flows, and it's refreshing to see LLMs absorb the ambiguity we used to manually script around. The hard part now isn't the bot's logic, but knowing where to draw the line between letting it improvise and keeping guardrails. Curious how others handle that balance. AI video generation
The shift from rigid NLU flows to LLM-driven conversation is exactly the liberation we needed—now it’s about guiding intent, not predicting every branch. I’ve been testing this approach with a few projects and the difference in iteration speed is night and day, check out https://3mf-to-stl.com
The shift from rigid NLU flows to LLM-driven conversation is exactly the liberation we needed—no more mapping every edge case. Curious how you’re handling the balance between creativity and guardrails, I’ve been testing a few approaches with https://ai-logo-generator.com
The shift from rigid NLU flows to LLM-driven conversations is exactly the breath of fresh air this space needed—now it’s about guiding intent rather than coding every branch. Curious how you’re handling context windows in longer sessions, I’ve been tinkering with a similar approach over at https://gputomine.com
The shift from rigid NLU flows to letting LLMs handle the messy middle is such a relief—it finally feels like we're designing conversations instead of decision trees. I've been using a similar approach to cut down on intent-slotting overhead, and the difference in iteration speed is night and day. Check out https://aibestfinder.com