V60

voyager60

Advancing the understanding of political bias in news media through cutting-edge LLM research and community-driven data collection

7-Point
Bias Classification Scale
3-Way
Perspective Analysis
AI-Powered
Bias Detection

Our Research Mission

We're pioneering research into how Large Language Models can detect, classify, and understand political bias in news media, while exploring the fascinating interplay between AI bias and human perspective.

The Challenge

Political bias in news media spans a spectrum from Far Left to Far Right, with subtle nuances that even humans struggle to detect consistently. News organizations exhibit ideological "spin" through framing, word choice, and story selection.

External rating services like AllSides and Media Bias/Fact Check categorize outlets, but can AI models reliably detect these biases? And how does an AI's own inherent bias affect its judgment?

Bias Spectrum

Far Left
Left
Lean Left
Center
Lean Right
Right
Far Right

Three Research Directions

Bias Classification

Training LLMs to classify news articles across the 7-point political bias spectrum, exploring how model bias affects detection accuracy.

Zero-shot prompting approaches
Fine-tuning on labeled datasets
Cross-model bias analysis

Perspective Generation

Generating biased summaries from different political perspectives, testing controlled abstractive summarization with ideological spin.

Left/Center/Right summaries
Multi-document synthesis
Perspective faithfulness metrics

Metacognitive Effects

Investigating whether performing bias-heavy tasks temporarily shifts an LLM's own political orientation and response patterns.

Political Compass testing
Before/after bias measurement
Persona adoption analysis

Central Hypothesis

"A language model with inherent political bias may more easily detect opposite-wing bias than same-wing bias, as contrasting viewpoints appear more salient against the model's default perspective."

Left-leaning model detects right bias easily
Right-leaning model detects left bias easily

Help Advance Research

Your contributions to our Ground News URL collection directly support this cutting-edge research

Data Collection

Every Ground News URL you submit helps build our comprehensive dataset of articles across the political spectrum, enabling robust bias classification training.

Submit URLs

Earn Rewards

Get compensated ৳0.50 for each valid Ground News article you contribute, while supporting important research into AI and media bias.

View Earnings

Research Impact

Media Literacy

Help readers identify bias and understand multiple perspectives on current events

AI Safety

Advance understanding of how AI models handle political content and maintain neutrality

Academic Research

Contribute to peer-reviewed research on LLM behavior and political bias detection