Uploaded November 2025 | Updated September 2026, 1 week ago

---

Large language models (LLMs) generate diverse responses, yet trial-and-error prompt engineering is inefficient for finding the ideal ones. What other levers exist? This talk introduces a principled approach to steering word sampling during generation, giving users and developers greater control to align outputs with tasks, goals and business needs.

Speaker:
Runyan Tan, Principal Consultant, Thoughtworks

thoughtworks.com/xconf