Filtir

Fact-check AI-Generated Text and Verify Claims

Free Trial

Description

Filtir is an AI-powered fact-checking tool available both as a web application and a ChatGPT Plugin. It is designed to address the issue of "hallucinations" or false factual claims often present in content generated by AI assistants like ChatGPT. Filtir works by meticulously analyzing a body of text to identify factual claims.

For each identified claim, the tool searches for public, auditable evidence to determine if the claim is supported. Rather than declaring claims as definitively true or false, Filtir presents the supporting evidence it finds, allowing users to make their own assessment. Claims lacking readily available public evidence are flagged as unsupported, enhancing the reliability of AI-assisted content creation.

Key Features

Fact-Checking Engine: Identifies factual claims within text.
Evidence Assessment: Searches for and evaluates public, auditable evidence for claims.
Claim Verification Status: Flags claims as verified, partially verified, or unsupported.
Source Provision: Displays supporting evidence and sources for verified claims.
ChatGPT Plugin Integration: Seamlessly fact-checks within the ChatGPT interface.
Web Application Access: Offers a standalone web app for fact-checking.
Support Score: Provides a summary score indicating the proportion of verifiable facts.

Use Cases

Verifying factual accuracy of AI-generated articles and reports.
Fact-checking content created with ChatGPT or similar AI assistants.
Ensuring reliability of information before publishing.
Identifying and mitigating AI hallucinations in text.
Cross-referencing claims against public evidence sources.

Frequently Asked Questions

How do you determine whether a factual claim is true?

Filtir does not arbitrate truth. Instead, it provides public, auditable evidence in support of factual claims found in the text. Claims where appropriate evidence cannot be readily identified are flagged as 'unsupported', meaning public evidence wasn't found, not necessarily that the claim is false.

Which sources of evidence do you consider acceptable? How do you know if a source is "trustworthy"?

The objective is to provide supportive evidence where it can be found. Currently, Filtir does not discard evidence based on reputation metrics; it trusts the user to assess the source's trustworthiness. This strategy may be revised based on user feedback.

Does Filtir sell my data?

No, Filtir does not sell user data or share content with third parties for marketing purposes, as detailed in their privacy and API data policies.