Natural Language Training: Best Practices for Bixby
summary
Shows a concrete workflow: load the movie agent capsule from GitHub into Bixby Studio, add training, compile to flip examples from "not learned" to "learned", and verify plans in the simulator and planner. Demonstrates 29 initial training examples, uses vocab files and prompts (a quiz) to reduce required examples, and finds that with four diverse examples Bixby generalizes to three new utterances. Uses the capsule interpreter report to surface issues (example: 32 learned, 2 illegal plans) and fixes annotations until the capsule works end-to-end.
AI-written summary · about
chapters (12)
- 0:00Introduction to NL training
- 3:41Key NL training concepts
- 7:15Demo in Bixby Studio
- 13:23Defining training sets
- 16:40Outers and continuations
- 20:04Using vocab effectively
- 25:45How many examples needed
- 27:09Iterative training cycle
- 29:00Prompt training explained
- 35:10Capsule interpreter report
- 41:39Troubleshooting not learned
- 43:28Final checklist
related talks
39mLet's Talk: A Focus on Natural Language and Conversational Dialogue with Bixby
40mTeaching Bixby Fundamentals: What You Need to Know
45mBest Practices from the Team That Built Bixby
34mConquer the World with Bixby: Building for Multiple Languages and Devices
49mBixby: A New Take on the Intelligent Assistant
41mBixby and the New Exponential Frontier of Intelligent Assistants
45mConversational Commerce with Bixby: The Next Big Thing in Assistants
01h 04mBuilding Expertise Into Generated Conversations