CHAPTERS
- 0:00 – 0:03
Defining an “LLM whisperer” role at Anthropic
The clip opens by framing the central question: what skills and habits make someone effective at eliciting strong behavior from large language models. This sets up the discussion as practical, day-to-day craft rather than abstract theory.
- •Positions the topic as a distinct skill: “LLM whisperer”
- •Focus on what it takes in practice at Anthropic
- •Signals that the answer will be behavior- and interaction-driven
- 0:03 – 0:33
Hands-on, empirical model interaction and experimentation
Amanda emphasizes that the core requirement is extensive, repeated interaction with models and careful observation of outputs. She describes the work as fundamentally empirical: build intuition by experimenting and learning the “shape” of model responses.
- •Spend lots of time interacting with models and reviewing outputs
- •Develop intuition for how models respond across prompts and contexts
- •Treat prompting/model-steering as an experimental, empirical discipline
- •Willingness to iterate and test variations to learn what changes behavior
- 0:33 – 0:39
Debugging misunderstandings: clarify intent, ask why, and refine prompts
When the model behaves unexpectedly, Amanda describes two approaches: ask the model to explain, or inspect the prompt to identify what caused the misunderstanding. She closes by noting this process reveals surprising depth in model behavior and makes the work compelling.
- •Explain the issue/concern to the model as clearly as possible
- •When outputs are unexpected, query the model’s reasoning (“ask it why”)
- •Diagnose what in the prompt led to misinterpretation
- •Iterate on wording and framing to better align model behavior
- •Model behavior can reveal “interesting depths,” making the work engaging
