How Verification Works
When you chat with any AI on GlyphChat, something special happens behind the scenes. Here’s how our GlyphSafe system verifies responses.
The GlyphSafe System
Step 1: AI Generates Response
Your chosen AI (ChatGPT, Claude, Gemini, etc.) creates a response to your question.
Step 2: Claim Extraction
Our system identifies individual claims within the response. For example:
- “Paris is the capital of France” → 1 claim
- “Water is made of hydrogen and oxygen” → 1 claim
Step 3: Concept Matching
Each claim is analyzed to find the key concepts:
- “Paris is the capital of France” → Paris, capital, France
Step 4: Knowledge Graph Check
We search our Glyphmine database (28,100+ concepts, 184,000+ relationships) to verify if these concepts are connected correctly.
Step 5: Confidence Score
Based on what we find, we assign a confidence score:
| Score Range | Meaning |
|---|---|
| 90-100% | Strong verification path found |
| 70-89% | Partial verification |
| 50-69% | Weak connection |
| 0-49% | Could not verify |
What This Means for You
Remember: Verification is a guide, not absolute truth. High confidence means we found supporting evidence in our knowledge base. Low confidence doesn’t necessarily mean the AI is wrong—it may just mean we couldn’t verify it.
Key Takeaways
- Trust responses with high confidence scores - These align with verified knowledge
- Low scores don’t mean the AI is wrong - Just that we couldn’t verify it
- Use verification as a guide - For important decisions, always double-check critical facts
The Glyphmine Database
Our verification is powered by Glyphmine, a knowledge database containing:
- 28,100+ concepts - From basic facts to specialized knowledge
- 184,000+ relationships - How concepts connect to each other
- Constantly expanding - We’re always adding more verified knowledge
Next Steps
- Learn to read verification badges
- Explore concepts with the Concept Explorer