Weight-Space Programming enables localized neural edits with up to 29.45-point gains
16 trajectories on Qwen models reveal WSP's strong localized improvements but inconsistent generalization and reliability
Stories per day · 30 days
2 RSS
Relatedcortex1evals1gemma1gradient-free1intra-weight-injection1llm1lora1memory-consolidation1mixture-of-experts1neural-networks1qwen1weights1weight-space-programming1
10 stories, newest first · one card per story
Stories tagged #model-editing
IWIBench tests gradient-free intra-weight injection by editor models
Astra and GPT-6.1 Sol lead editors of Gemma 3 4B IT; structured-response edits resisted every attempt.
Astra AI Post on Qwen Weights Misleading About Capabilities
Astra model experiments on Qwen use code execution to edit weights rather than reading them directly for improvements.
1
Cortex: Consolidating Context into Parametric Memory
Cortex writes context gains into model weights, recovering 88.9% after 50 edits on a 14B model.
Certified Mechanistic Edits: Behavioral Guarantees for Skill Removal and Preservation
Researchers certify that mechanistic edits provably remove one neural network skill while preserving another, replacing test-only validation with formal behavioral guarantees.
Imprint Reader: From Weight-Update Readout to Behavioral Intervention
An Imprint Reader describes frozen weight updates and guides edits that raise refusal and tool-use scores.
A Chosen Future Can Still Be Rewritten: Causal Writability in Video Models
Study shows video models often learn correct physics but fail to use it; low-dimensional 'causal writability' edits can restore correct motion.
Disentangling Representation Evolution in Transformers through Directional Decomposition
Decomposes transformer representation updates into parallel and perpendicular components, linking geometry to editing robustness and better pretraining.
Disentangling Representation Evolution in Transformers through Directional Decomposition
Researchers decompose transformer updates into parallel and perpendicular components, linking representation geometry to editing robustness, compression diagnosis, and training interventions.
The information geometry of large language models is shared, learned, and controllable
Fisher-Rao output geometry is shared across LLM architectures and enables minimum-disturbance interventions that transfer across prompts for steering and editing.