2410.04332
arxiv.org · 7,905 words · saved by 1 readers
N/A
Preprint G RADIENT ROUTING : M ASKING G RADIENTS TO L O - CALIZE C OMPUTATION IN N EURAL N ETWORKS Alex Cloud∗ , Jacob Goldman-Wetzler∗ , Evžen Wybitul∗ , Joseph Miller∗ ML Alignment and Theory Scholars (MATS) Alexander Matt Turner arXiv:2410.04332v2 [cs.LG] 29 Nov 2024 A BSTRACT…
related reading
- Gradient Routing: Masking Gradients to Localize Computation in Neural Networks — AI Alignment Forumalignmentforum.org
- Gradient Routing: Masking Gradients to Localize Computation in Neural Networks — LessWronglesswrong.com
- Circuit Tracing: Revealing Computational Graphs in Language Modelstransformer-circuits.pub
- Selective modularity: a research agenda — LessWronglesswrong.com
- NL.pdfabehrouz.github.io
- [2510.13786] The Art of Scaling Reinforcement Learning Compute for LLMsarxiv.org
- [2512.05648] Beyond Data Filtering: Knowledge Localization for Capability Removal in LLMsarxiv.org
- 2310.01405.pdfarxiv.org
- 2310.01405arxiv.org
- Language Model Circuits Are Sparse in the Neuron Basis | Transluce AItransluce.org
- Modular Pretraining Enables Access Controlalignment.anthropic.com
- [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org