Dialogue design evolves with LLMs, not dies, says Slang AI exec

In a podcast discussion, Rebecca Evanhoe of Slang AI argued that dialogue design is evolving, not dying, shifting from traditional NLU to context engineering with large language models (LLMs). Slang AI employs a hybrid system for restaurant voice AI, using LLMs for tasks like menu inquiries while retaining deterministic rules for core business logic and API integrations to manage costs and reduce hallucinations. The company is developing new evaluation methods for LLMs, including using LLMs as judges and a "Brain Trust" approach with human annotators to train specialized evaluation models. Evanhoe also highlighted the unique challenges of voice AI, such as environmental noise and latency, and how AI can provide prescriptive business insights to restaurants by analyzing real customer interaction data. AI

IMPACT Dialogue design is adapting to LLMs, focusing on context engineering and new evaluation methods, offering prescriptive business insights for industries like restaurants.

RANK_REASON Podcast discussion about evolving dialogue design practices with LLMs.

Read on dev.to — LLM tag →

• Brain Trust
• Conversations with Things
• LLM
• NLU
• OpenTable
• Rasa
• Rebecca Evanhoe
• Slang AI
• The Dialogue Architects

• product
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →