D-STEER - Preference Alignment Techniques Learn to Behave, not to Believe -- Beneath the Surface, DPO as Steering Vector Perturbation in Activation Space

Documenti analoghi

• TRACEALIGN -- Tracing the Drift: Attributing Alignment Failures to Training-Time Belief Sources in LLMs
  di: Das, Amitava, et al.
  Pubblicazione: (2025)
• MENTIS: What Belief Changes Under Alignment? Measuring Multi-Scale Latent Torsion in Language Models
  di: Saha, Partha Pratim, et al.
  Pubblicazione: (2026)
• AlignGuard-LoRA: Alignment-Preserving Fine-Tuning via Fisher-Guided Decomposition and Riemannian-Geodesic Collision Regularization
  di: Das, Amitava, et al.
  Pubblicazione: (2025)
• SPINAL -- Scaling-law and Preference Integration in Neural Alignment Layers
  di: Das, Arion, et al.
  Pubblicazione: (2026)
• Neural FOXP2 -- Language Specific Neuron Steering for Targeted Language Improvement in LLMs
  di: Saha, Anusa, et al.
  Pubblicazione: (2026)