Skip to content
All projects
ResearchIn progress

Steering Truth in LLMs

Feature-engineering truth and modality directions in LLM activations, then steering with XGBoost to beat linear-probe and contrastive-mean-difference baselines.

Researcher, Trustworthy Robotics Lab · 2026 · Trustworthy Robotics Lab

Source

Demo walkthrough coming soon

Why I built it

draft

Truth and other high-level concepts show up as directions in a model's activations, but the usual ways to recover them (a linear probe, or the contrastive mean difference between true and false statements) are blunt. I treated it as a feature-engineering problem: find richer directions, then steer with XGBoost to move behavior further than the linear baselines could. On gemma-2-2b, a direction I could decode at ~99% accuracy still wasn't one I could steer on.

Stack

  • Python
  • PyTorch
  • gemma-2-2b
  • XGBoost
  • NCSA DeltaAI (GH200)
ML/AIApplied-Research
Back to all projects