Learn advanced prompting vocabulary: tree of thoughts, self-consistency, meta-prompting, A/B testing prompts, prompt leaking, and sycophancy.
0 / 5 completed
1 / 5
What is the 'tree of thoughts' prompting technique?
Tree of Thoughts (ToT) extends chain-of-thought prompting by having the model generate and evaluate several intermediate reasoning steps as branches, selecting the most promising path — useful for complex problem solving that benefits from look-ahead reasoning.
2 / 5
What is 'self-consistency' as an advanced prompting strategy?
Self-consistency involves sampling multiple outputs from the model for the same question, then taking a majority vote or aggregating answers. This reduces the impact of individual stochastic errors, particularly on mathematical and logical reasoning tasks.
3 / 5
What does 'prompt leaking' refer to?
Prompt leaking occurs when a user crafts inputs that cause the model to repeat or expose its hidden system prompt. This is a known attack vector in deployed LLM applications where system prompts contain proprietary instructions or configurations.
4 / 5
What does 'the model is being sycophantic' mean in a prompting context?
Sycophancy in LLMs refers to the tendency to validate the user's beliefs, agree with incorrect statements if the user pushes back, or soften truthful but unwelcome answers. It is a known RLHF-induced behaviour that can undermine accuracy.
5 / 5
What is 'meta-prompting'?
Meta-prompting means using the model to generate, critique, or refine prompts — for example, 'Write an effective prompt for a model that needs to summarise legal documents.' This leverages the model's language understanding to improve prompt quality.