Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Gradient Atoms decomposes per-document training gradients into sparse components via dictionary learning in a preconditioned eigenspace.method 95%
• Among 500 discovered atoms, the highest-coherence ones recover interpretable task-type behaviors without any behavioral labels.result 90%
• Applying discovered atoms as weight-space perturbations produces large, controllable shifts in model behavior.result 90%
• The method requires no query-document scoring stage and scales independently of the number of query behaviors of interest.technical 85%
• Existing TDA methods are supervised, requiring a query behavior and scoring every training document against it.limitation 90%
• Per-document framing of training data attribution is fundamentally mismatched to how fine-tuning actually works.technical 80%
• Steering vectors could have unintended side effects on model performance.limitation 70%
• The method may not generalize to all model architectures or training regimes.limitation 70%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card