Create a summary markdown table of the LLM prompting techniques covered here https://www.forbes.com/sites/lanceeliot/2024/05/09/the-best-prompt-engineering-techniques-for-getting-the-most-out-of-generative-ai/. Be certain to include in this table an approximate token usage range for each method (with one-shot prompting being very low and chain-of-thought possibly being very high).
TechniqueDescriptionApprox. Token Usage Range
Browbeating PromptsUses assertive or demanding language to push AI for better responses.Medium (~100-300)
Catalogs/Frameworks PromptingProvides a structured list or framework to guide AI’s response.Medium (~100-300)
Certainty/Uncertainty PromptingSpecifies confidence level desired in AI’s response (e.g., certain or exploratory).Low (~50-100)
Chain-of-Density (CoD) PromptingIteratively increases detail density in responses for concise yet rich outputs.High (~300-1000)
Chain-of-Feedback (CoF) PromptingIncorporates iterative feedback loops to refine AI outputs.High (~300-1000)
Chain-of-Thought (CoT) PromptingEncourages step-by-step reasoning to improve problem-solving accuracy.Very High (~1000+)
CoT Factored Decomposition PromptingBreaks down reasoning into sub-steps for complex tasks, building on CoT.Very High (~1000+)
Chain-of-Verification (CoV) PromptingPrompts AI to verify each step of its reasoning to reduce errors.High (~300-1000)
Conversational PromptingEngages AI in a dialogue-like exchange for natural, iterative responses.Medium (~100-300)
DeepFakes to TrueFakes PromptingGuides AI to distinguish or generate realistic vs. deceptive content.Medium (~100-300)
Directional Stimulus Prompting (DSP)Provides hints or directional cues to steer AI’s response.Low (~50-100)
Disinformation Detection PromptingInstructs AI to identify and flag potential misinformation in responses.Medium (~100-300)
Emotionally Expressed PromptingUses emotional tone (e.g., excitement, urgency) to influence AI’s response style.Low (~50-100)
End-Goal PromptingSpecifies the desired outcome upfront to focus AI’s response.Low (~50-100)
Essay-Compression PromptingAsks AI to summarize or compress lengthy content into concise forms.Medium (~100-300)
Fair-Thinking PromptingEncourages balanced, unbiased responses from AI.Low (~50-100)
Flipped Interaction PromptingReverses roles, asking AI to question the user or simulate a scenario.Medium (~100-300)
Generating Prompts via Generative AIUses AI to craft prompts for itself, automating prompt creation.Medium (~100-300)
Illicit/Disallowed PromptingAttempts to bypass AI restrictions (ethically questionable).Low (~50-100)
Imperfect PromptingEmbraces flawed prompts, refining through iteration rather than striving for perfection initially.Low (~50-100)
Importing Text as PromptingFeeds external text (e.g., documents) into AI as context for responses.High (~300-1000)
One-Shot PromptingProvides a single example to guide AI’s response (not explicitly listed but implied in context).Very Low (~10-50)