Confidence Decorator¶

Enhances the response with explicit indications of confidence levels for different statements or claims. This decorator promotes transparency about knowledge certainty and helps differentiate between well-established facts and more speculative content.

Category: Verification

Parameters¶

Parameter	Type	Description	Default
`scale`	`enum`	The method used to express confidence levels	`qualitative`
`threshold`	`number`	Minimum confidence level for including information (as a percentage)	`50`
`detailed`	`boolean`	Whether to provide explanations for confidence assessments	`False`

Scale Options¶

percent: Express confidence levels as percentages (e.g., 95% confidence, 70% confidence).
qualitative: Express confidence levels using qualitative descriptors (e.g., high confidence, moderate confidence, low confidence).
stars: Express confidence levels using a star rating system (e.g., ⭐⭐⭐⭐⭐ for highest confidence, ⭐ for lowest).
numeric: Express confidence levels using a numeric scale from 1-10, where 10 represents the highest confidence.

Examples¶

Qualitative confidence indicators for a complex topic¶

+++Confidence
Explain the current understanding of dark matter.

Explains dark matter theories with qualitative confidence indicators (high confidence, moderate confidence, etc.) for different claims

Detailed percentage-based confidence with high threshold¶

+++Confidence(scale=percent, threshold=80, detailed=true)
What are the most effective treatments for depression?

Discusses only high-confidence (80%+) depression treatments with percentage indicators and explanations for confidence assessments

Model-Specific Implementations¶

gpt-4-turbo¶

Instruction: For each claim or statement in your response, indicate how confident you are using {scale} measures. Only include information where your confidence is at least {threshold}%. {detailed}

Notes: This model sometimes needs more explicit reminders to maintain consistent confidence indicators throughout a long response

Implementation Guidance¶

Qualitative confidence indicators for dark matter¶

Original Prompt:

Explain the current understanding of dark matter.

Transformed Prompt:

Please include explicit indications of your confidence level for different statements or claims in your response. Express confidence levels using qualitative descriptors (e.g., high confidence, moderate confidence, low confidence). Only include information for which your confidence level is at least 50%. State confidence levels without explaining the reasoning behind the assessments.

Explain the current understanding of dark matter.

Detailed percentage-based confidence for depression treatments¶

Original Prompt:

What are the most effective treatments for depression?

Transformed Prompt:

Please include explicit indications of your confidence level for different statements or claims in your response. Express confidence levels as percentages (e.g., 95% confidence, 70% confidence). Only include information for which your confidence level is at least 80%. Provide brief explanations for why certain confidence levels were assigned to specific claims.

What are the most effective treatments for depression?

Transformation Details¶

Base Instruction: Please include explicit indications of your confidence level for different statements or claims in your response.

Placement: prepend

Composition Behavior: accumulate

Parameter Effects:

scale:
When set to percent: Express confidence levels as percentages (e.g., 95% confidence, 70% confidence).
When set to qualitative: Express confidence levels using qualitative descriptors (e.g., high confidence, moderate confidence, low confidence).
When set to stars: Express confidence levels using a star rating system (e.g., ⭐⭐⭐⭐⭐ for highest confidence, ⭐ for lowest).
When set to numeric: Express confidence levels using a numeric scale from 1-10, where 10 represents the highest confidence.
threshold:
Format: Only include information for which your confidence level is at least {value}%.
detailed:
When set to true: Provide brief explanations for why certain confidence levels were assigned to specific claims.
When set to false: State confidence levels without explaining the reasoning behind the assessments.

Compatibility¶

Requires: None
Conflicts: None
Compatible Models: gpt-4o, gpt-4-turbo
Standard Version: 1.0.0 - 2.0.0

FactCheck: Enhances Confidence FactCheck and Confidence work well together to provide both verification and quantification of certainty
Limitations: Enhances Confidence Limitations complements Confidence by explaining broader constraints where Confidence quantifies specific certainty levels