Conversational Bots: Panel Discussion
summary
Panelists debate why chatbots stumble and give concrete tactics: training data and the long tail force firms to constrain the domain or suffer noisy models; start with focused tasks like password reset, build intent classifiers (even Naive Bayes), use multimodal inputs and buttons, and keep humans in the loop for reinforcement. Recommend mapping backend APIs into semantic parsing representations, share anonymized corpora via industry consortia, and cite working examples such as Statsbot and Polly as winning, pragmatic patterns.
AI-written summary · about
chapters (5)
- 0:00Start
- 6:28Advocating Multimodal Input
- 23:06The Very First Stage That You See When You'Re Building a Chat Bot
- 33:19What Not To Do
- 35:16Use Humans in the Loop
related talks
50mChatbots state of the art and next challenges
10mMastering LLM Chatbot Testing: Metrics, Methods and Mistakes to Avoid
45mPanel Talk
17mArchitecting LLM-Based Chatbots using free open-source LLMs
39mHow Chatty Leads Builds AI Agents Businesses Actually Want
49mBuilding multilingual and empathetic bots
42mKeynote: Designing LLM Native systems
25mConversational AI: Sales and Customer Service Chatbot Use Cases for Small and Medium Sized Business