Generative AI Insights
Crafting Exceptional Conversational Experiences with AI requires more than just deploying an LLM (Large Language Model). As we continue to push the boundaries of generative artificial intelligence, it's essential to acknowledge that building a high-quality chatbot requires strategic planning and execution.
1. Leverage Scripted Responses
While LLMs are incredibly powerful, they're better utilized in controlled environments rather than as a standalone talking box. By leveraging scripted responses, you can save money while delivering a better user experience. A simple decision tree with semantic reasoning for routing to serve common user journeys can lead to a more effective product.
This strategy involves using pre-written responses to engage users and provide a seamless conversational experience. Scripted responses can be particularly effective when combined with decision trees, allowing you to create a more personalized and interactive experience for your users.
2. Craft Amazing Conversation Starters
A well-crafted conversation starter can significantly enhance the user experience. We developed an entirely separate inference system from our core dialog system to use summaries of previous chats, memories, and recent app actions to initiate engaging conversations. This approach prevents your AI from producing generic responses like "Hi! How can I assist you today?"
By using conversation starters, you can set the tone for a more personalized and engaging experience. This strategy involves using pre-written messages or prompts to encourage users to engage with your chatbot in a more meaningful way.
3. Route Messages to Multiple Models
Route messages in the same thread to two to six models with orthogonal capabilities instead of relying on a single model. This strategy can lead to dramatically better outcomes over multi-turn conversations. Choose your models wisely, and consider running split tests for advanced prompting.
This approach involves using multiple AI models to engage users and provide a more comprehensive experience. By routing messages to different models, you can create a more dynamic and interactive conversation that caters to the needs of your users.
4. Use Scripted Responses Effectively
Scripted responses are a crucial aspect of building high-quality chatbots. By using a smaller model to infer semantics about user input and routing to pre-written responses, you can create a more engaging experience while saving costs. This approach is particularly effective when combined with decision trees.
5. Have a Clear Metric to Judge AI Output
It's essential to have a clear metric to judge AI output, especially after a 100-turn conversation across multiple sessions and user personas. Use a feedback loop with a Likert score or simple ELO score to choose between variants and see what users find engaging or useful in chat.
6. Be Aware of the Quality of AI-to-AI Chats
During user testing, we repeatedly saw the blank canvas problem – users didn't know what to type to chat. We added a "magic wand" to offer three AI-generated messages in the user's voice, which solved short-term user friction but led to faster churn. AI-to-AI chat degrades quickly into nonsense within a few turns.
Conclusion
Crafting exceptional conversational experiences with AI requires more than just deploying an LLM. By leveraging scripted responses, crafting amazing conversation starters, routing messages to multiple models, using scripted responses effectively, having a clear metric to judge AI output, and being aware of the quality of AI-to-AI chats, you can build a high-quality chatbot that engages and delights users.
About the Author
Siamak Freydoonnejad is co-founder of Campfire. With extensive experience in building conversational interfaces, Siamak has developed innovative strategies for creating exceptional chatbots that engage and delight users. His expertise lies in leveraging scripted responses, conversation starters, and multiple models to create a more personalized and interactive experience.