V7 Go Update: Visualize AI citations and information sources in PDFs
A strategic shift towards more transparent, accurate, and reliable AI. See exactly where your AI models draw their conclusions and reasoning from.
July 15, 2024
5
mins read
As AI systems become more sophisticated, they've also become harder to understand and control. This leads to the "black box" problem where we have to constantly question the reliability of AI outputs. That’s why trust and accuracy have become key concerns for fintech companies, law firms, or healthcare institutions who try to implement AI.
When the stakes are high and there is no room for mistakes. It's not just about getting an answer. It's about understanding where that answer came from and why it's correct or not.
To solve this problem, we developed AI citations.
AI citations powered by visual grounding allow users to see exactly where an AI model has drawn its reasoning from within source documents. Here's a breakdown of its functionality:
The AI automatically highlights areas in your documents that were used as information sources.
Users can click on sentences in the AI's response to navigate to the relevant citation in the input document.
The system provides a visual representation of the AI's "thought process," allowing for instant verification of generated responses.
This process creates a clear, auditable trail from user input to AI output.
Let’s imagine that we're reviewing a set of financial reports, and you ask an AI to summarize key points.
With visual grounding, you're not just getting a summary – you're getting a summary with footnotes.
The key benefits:
Fewer hallucinations. By requiring the AI to back up its answers with direct citations, we greatly reduce the chances of it "making things up.”
Faster quality checks. Tasks that used to take hours of manual checking can now be done in minutes. The system directly points to the relevant information.
Increased trust. For professionals working with clients, being able to show the sources of AI-generated insights boosts confidence and credibility.
The implications of visual grounding extend far beyond simple fact-checking. Here are just a few use cases that can benefit from this game-changing technology:
Legal document review. Lawyers can rapidly verify AI-extracted clauses against source contracts, dramatically speeding up due diligence processes.
Financial audits. Accountants can trace AI-flagged discrepancies directly to relevant entries in financial statements.
Regulatory compliance. Fintech companies can demonstrate clear audit trails for AI-assisted decision-making processes to regulators.
How to use it:
To leverage AI citations in your V7 Go workflows:
Toggle Grounding in your AI property setup
Inspect a selected entity and click the lightbulb icon.
Click on sentences to navigate to the relevant citation.
This simple process transforms complex document analysis tasks into intuitive, verifiable interactions with your AI.
Technical deep dive:
When activated, Grounding adds approximately 20% to the token cost of a query. However, this additional cost often pays for itself in improved accuracy and reduced review time.
The feature currently works with PDF inputs and text outputs only.
Grounding coordinates are exposed via the API, allowing for deep integration into existing workflows and applications.
Explore V7’s Auto-Track feature for faster video annotation. Use automated instance segmentation and object tracking across frames to reduce delivery times for video labeling projects.
Discover how V7's new features for medical image segmentation can transform your radiology projects with advanced 3D visualization, custom layouts, improved segmentation models, and more.