AI Training Data Contribution

The Sampling screen serves as your control center for managing AI sampling requests. When the AI system identifies opportunities to learn from your interactions, you maintain complete control over what data can be used for model improvement.

Overview

Sampling in systemprompt represents a privacy-first approach to AI improvement. When the language model encounters unique or valuable interactions, it may request permission to use anonymized versions of these exchanges for training purposes. You have complete control over these requests.

Key Principles

• User Control: Nothing is sampled without explicit permission
• Transparency: Clear explanation of what will be sampled
• Privacy Protection: All personal data is anonymized
• Opt-in Only: Sampling is never automatic
• Revocable: Change your mind anytime

Understanding Sampling Requests

What Triggers Sampling

The AI may request sampling when:

• Novel Interactions: Unique tool combinations or workflows
• Error Recovery: Successful problem resolution patterns
• Complex Queries: Multi-step operations that worked well
• Edge Cases: Unusual but valid use cases
• Feedback Loops: Corrections that improved results

What's Included

Sampling requests show exactly what would be shared:

Included

• Tool sequences used
• Command patterns
• Error recovery steps
• Workflow structures
• Success indicators

Excluded

• Personal identifiers
• Specific data values
• Private content
• Authentication details
• Server URLs

The Sampling Interface

Request Display

Each sampling request shows:

1. Request Title: Clear description of the interaction type
2. Preview: Anonymized version of the data
3. Purpose: How this helps improve the AI
4. Data Scope: Exactly what would be shared
5. Expiry: How long you have to decide

Review Process

Step 1: Notification

• Badge on Sampling tab
• Optional push notification
• Non-intrusive alert

Step 2: Review

• Full preview of anonymized data
• Highlighted sections
• Explanation of value
• Privacy guarantees

Step 3: Decision

• Approve: Contribute to improvement
• Deny: No data shared
• Modify: Adjust what's shared
• Defer: Decide later

Making Decisions

Approval Considerations

Before approving, consider:

• Value: Does this help others?
• Privacy: Is data properly anonymized?
• Comfort: Are you okay sharing this?
• Uniqueness: Is this interaction special?

Approval Options

Full Approval

• Share complete interaction pattern
• Maximum value for improvement
• Fully anonymized

Partial Approval

• Select specific parts to share
• Remove sensitive sections
• Maintain control

Conditional Approval

• Set terms for usage
• Time-limited sharing
• Specific improvement goals

Denial Options

Simple Denial

• One tap to decline
• No explanation needed
• No negative consequences

Deny with Feedback

• Explain privacy concerns
• Suggest improvements
• Help refine requests

Privacy Controls

Anonymization Process

Before any data is shared:

1. Identifier Removal: All personal data stripped
2. Value Replacement: Real values become placeholders
3. Pattern Preservation: Structure maintained
4. Context Generalization: Specific becomes general
5. Review: Final check before sharing

Example Anonymization

Original

"Create GitHub issue #1234 in mycompany/api-backend 
about payment timeout for customer john@email.com"

Anonymized

"Create GitHub issue [ID] in [REPO] 
about [ERROR_TYPE] for customer [EMAIL]"

Your Rights

• Full Transparency: See exactly what's shared
• Withdrawal: Revoke permission anytime
• Deletion: Request removal of shared data
• Audit: Track what you've approved
• Control: Granular permission settings

Sampling Settings

Global Preferences

Configure default behaviors:

Auto-Handling

• Always approve certain types
• Always deny certain types
• Ask every time (default)

Notification Preferences

• Push notifications on/off
• Batch notifications
• Priority levels
• Quiet hours

Privacy Levels

• Maximum: Very strict anonymization
• Balanced: Standard anonymization
• Minimal: Basic anonymization

Category Controls

Set preferences by type:

• Error Patterns: Usually valuable to share
• Tool Workflows: Help others discover features
• Voice Commands: Improve recognition
• Recovery Steps: Help prevent issues

Benefits of Participation

Community Impact

Your contributions help:

• Improve AI Understanding: Better responses for everyone
• Enhance Tool Discovery: Others find useful workflows
• Reduce Errors: Learn from resolved issues
• Expand Capabilities: Enable new features

Personal Benefits

• Better AI Performance: Improved responses
• Feature Priority: Influence development
• Early Access: Preview new capabilities
• Recognition: Contributor badge (optional)

Managing History

Sampling History

View your contribution history:

• Approved Requests: What you've shared
• Denied Requests: What you declined
• Pending Requests: Awaiting decision
• Expired Requests: No longer available

Actions on History

For each historical item:

• View Details: See what was shared
• Revoke Permission: Remove from training
• Download Data: Get copy of contribution
• Report Issue: Flag concerns

Best Practices

When to Approve

Consider approving when:

• Interaction was particularly clever
• You solved a complex problem
• Pattern could help others
• No sensitive data involved

When to Deny

Consider denying when:

• Working with sensitive projects
• Unsure about anonymization
• Interaction too specific
• Privacy concerns exist

Optimal Participation

1. Review Carefully: Take time to understand
2. Start Small: Approve simple cases first
3. Build Trust: See how system works
4. Provide Feedback: Help improve process
5. Stay Informed: Check history regularly

Advanced Features

Bulk Management

Handle multiple requests efficiently:

• Select Multiple: Batch approve/deny
• Filter by Type: Focus on categories
• Quick Actions: Swipe gestures
• Smart Grouping: Similar requests together

Custom Rules

Create automation rules:

IF interaction_type = "error_recovery"
AND no_sensitive_data = true
THEN auto_approve

Analytics

Track your impact:

• Contribution Score: Overall impact metric
• Category Breakdown: Where you help most
• Trend Analysis: Contribution patterns
• Community Rank: Optional leaderboard

FAQ

Q: Can I change my mind after approving? A: Yes, you can revoke permission anytime from History.

Q: What if I accidentally approve sensitive data? A: Use emergency revoke in History. Data is queued before use.

Q: Do I need to participate? A: No, sampling is completely optional.

Q: How long is data retained? A: According to our data retention policy, typically 24 months.

Q: Can I see how my data was used? A: Yes, through the contribution tracking system.

Troubleshooting

Common Issues

Not Receiving Requests

• Check notification settings
• Verify sampling is enabled
• Use more unique workflows

Can't Approve/Deny

• Check internet connection
• Update app version
• Clear cache and retry

Concerns About Privacy

• Review anonymization examples
• Contact privacy team
• Adjust privacy level settings

Help improve systemprompt while maintaining complete control over your data!