AI UX is Broken, Not the Model

Source Domain: www.startuphub.ai

Here’s a summary of Mike Christensen’s presentation on “Why Your AI UX Is Broken (and It’s Not the Model’s Fault)” using an unordered list with 6 key points:

AI UX Issues are More Architectural Than Model-Based: The problems in AI user experience (UX) often stem from flawed infrastructure and architecture rather than from the AI models themselves. This misconception needs addressing to improve AI applications.
Default Architecture: HTTP Streaming: The most common approach for AI chats involves HTTP streaming where a user sends messages to a server, and the server processes these using a large language model (LLM), then sends a stream back. This method has several issues.
Problems with HTTP Streaming: The inherent problems with this setup include:
- Resilient Delivery: AI streams are lost when disconnected, breaking user experience.
- Continuity Across Surfaces: Difficulties maintaining seamless experiences across multiple devices.
- Live Control: Challenges in providing real-time feedback or allowing user control over AI outputs.
Solution: Decoupling and Durable Sessions: A more robust approach involves decoupling the agent layer from the client layer using durable sessions:
- Resilient Delivery: Sessions store events allowing users to reconnect and resume where they left off.
- Continuity Across Surfaces: Users can subscribe to the same session from different devices for consistent experience.
- Live Control: Better manage requests to steer or interrupt AI’s output for improved user control.
Ably’s AI Transport: This tool from Ably provides a real-world implementation of decoupling and durable sessions. It uses publish-subscribe channels for reliable, asynchronous communication, ensuring control over AI tasks and smoother UX.
Improved AI UX through Better Infrastructure: By shifting from HTTP streaming to a durable sessions model, developers can create more resilient, multi-device, and user-control-friendly AI applications leading to better user experiences.