This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
ApartSprints
Women in AI Safety Hackathon
679781551b57b97e23660edd
Women in AI Safety Hackathon
March 10, 2025
Accepted at the 
679781551b57b97e23660edd
 research sprint on 

HalluShield: A Mechanistic Approach to Hallucination Resistant Models

Our project tackles the critical problem of hallucinations in large language models (LLMs) used in healthcare settings, where inaccurate information can have serious consequences. We developed a proof-of-concept system that classifies LLM-generated responses as either factual or hallucinated. Our approach leverages sparse autoencoders (GoodFire’s Ember) trained on neural activations from Meta Llama 3. These autoencoders identify monosemantic features that serve as strong indicators of hallucination patterns. By feeding these extracted features into tree-based classification models (XGBoost), we achieved an impressive F1 score of 89% on our test dataset. This machine learning approach offers several advantages over traditional methods and LLM as a judge. First, it can be specifically trained on in-domain datasets (eg: medical) for domain-specific hallucination detection. Second, the model is interpretable, showing which activation patterns correlate with hallucinations and acts as a post-processing layer applied to LLM output.

By 
Tanzeel Shaikh, Naiyarah Hussain, Samuel Nihoul, Sonal Joshi, Lucia de la Torre
🏆 
4th place
3rd place
2nd place
1st place
 by peer review
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

This project is private