What it does
GoodHead streamlines the consumption of unbiased news to keep the public safely informed, one paragraph at a time.
In today’s fast-paced world, staying informed shouldn’t mean sacrificing time or clarity.
GoodHead's fine-tuned model for bias analysis scrapes news articles from various outlets and then outlines the rationale behind why the given paragraph may or may not be biased.
GoodHead is your essential co-pilot for navigating the modern news landscape with efficiency and safety.
How we built it
Microsoft Bing News Search API
Fine-tuned LLM via Intel AI
Uploaded to Hugging Face at rizzware/Llama-3.2-1B-Instruct-Bias
Posted on Twitter!
Groq for generation of additional features within training dataset
Challenges we ran into
Data, or lack thereof. There exists less than 5 datasets such that they contain text snippets of news articles alongside annotations of their biases.
Even with enough data, training an LLM on an annotated dataset is not the same as training a classification model which is usually employed for tasks such as bias detection.
GoodHead is for bias analysis - to inform our users of the rationale behind why a certain paragraph may or may not exhibit some degree of bias is more impactful than simply lettings users know that we do or don't detect bias.
Accomplishments that we're proud of
Completely solo - team of one, I deployed a model to Hugging Face and I finished a deploy-ready full-stack application!
What we learned
Know exactly what your MVP is. Over-engineering is a hacker's worst enemy. Why do you need Auth0 when - at demo-time - only 1 user will be using the platform (the PC being used for demoing)? That could be 30 minutes to 1 hour on a core feature.
What's next for GoodHead
GreatHead.
Log in or sign up for Devpost to join the conversation.