Residual Paving decomposes selective refusal editing into an early-layer router for intervention decisions and later-layer residual experts for edits, with oracle routing showing that learned route selectivity is the primary bottleneck across six backbones.
hub
Improving instruction-following in language models through activation steering.arXiv preprint arXiv:2410.12877
13 Pith papers cite this work. Polarity classification is still indexing.
13
Pith papers citing it
hub tools
citation-role summary
background 2
citation-polarity summary
verdicts
UNVERDICTED 13roles
background 2polarities
background 2representative citing papers
GCAD reduces coherence drift from -18.6 to -1.9 and raises turn-10 trait expression from 78.0 to 93.1 in persona-steering tasks by using gated attention-delta interventions from system prompts.
Instruction tuning makes late-layer computation depend more on the model's own post-trained upstream state than on base-model upstream state, producing a consistent +1.68 logit interaction effect across five model families.
RACC defines six representation-aware coverage criteria that score jailbreak test suites by measuring activation of safety concepts extracted from LLM hidden states on a calibration set.
Activation steering produces synthetic safety-violating data that improves downstream classifiers over prompting on most tested concepts when a harmonic mean of alignment, coherence, and diversity is optimized.
DMI-Lib delivers 0.4-6.8% overhead for offline batch LLM inference and ~6% for moderate online serving while exposing rich internal signals across backends, cutting latency overhead 2-15x versus prior observability baselines.
LLMs implicitly plan answer positions during MCQ generation, as shown by predictive signals in hidden representations and controllable shifts via activation steering.
LLMs implement a second-order confidence architecture where the PANL activation encodes both error likelihood and the ability to correct it, beyond verbal confidence or log-probabilities.
Language models deploy multidimensional internal confidence representations and threshold-based policies to control abstention behavior, with causal support from activation steering experiments.
Introduces IFBench benchmark with 58 new constraints and demonstrates RLVR training improves generalization of language models to unseen verifiable output constraints.
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
MemOS introduces a unified memory management framework for LLMs using MemCubes to handle and evolve different memory types for improved controllability and evolvability.
ReGA uses safety-critical representations to guide abstraction in model-based analysis, enabling scalable detection of harmful LLM inputs with reported AUROC of 0.975 at prompt level.
citing papers explorer
-
Residual Paving: Diagnosing the Routing Bottleneck in Selective Refusal Editing
Residual Paving decomposes selective refusal editing into an early-layer router for intervention decisions and later-layer residual experts for edits, with oracle routing showing that learned route selectivity is the primary bottleneck across six backbones.
-
Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions
GCAD reduces coherence drift from -18.6 to -1.9 and raises turn-10 trait expression from 78.0 to 93.1 in persona-steering tasks by using gated attention-delta interventions from system prompts.
-
Instruction Tuning Changes How Upstream State Conditions Late Readout: A Cross-Patching Diagnostic
Instruction tuning makes late-layer computation depend more on the model's own post-trained upstream state than on base-model upstream state, producing a consistent +1.68 logit interaction effect across five model families.
-
RACC: Representation-Aware Coverage Criteria for LLM Safety Testing
RACC defines six representation-aware coverage criteria that score jailbreak test suites by measuring activation of safety concepts extracted from LLM hidden states on a calibration set.
-
Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection
Activation steering produces synthetic safety-violating data that improves downstream classifiers over prompting on most tested concepts when a harmonic mean of alignment, coherence, and diversity is optimized.
-
Enabling Performant and Flexible Model-Internal Observability for LLM Inference
DMI-Lib delivers 0.4-6.8% overhead for offline batch LLM inference and ~6% for moderate online serving while exposing rich internal signals across backends, cutting latency overhead 2-15x versus prior observability baselines.
-
Do Large Language Models Plan Answer Positions? Position Bias in Multiple-Choice Question Generation
LLMs implicitly plan answer positions during MCQ generation, as shown by predictive signals in hidden representations and controllable shifts via activation steering.
-
How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signals
LLMs implement a second-order confidence architecture where the PANL activation encodes both error likelihood and the ability to correct it, beyond verbal confidence or log-probabilities.
-
Causal Evidence that Language Models use Confidence to Drive Behavior
Language models deploy multidimensional internal confidence representations and threshold-based policies to control abstention behavior, with causal support from activation steering experiments.
-
Generalizing Verifiable Instruction Following
Introduces IFBench benchmark with 58 new constraints and demonstrates RLVR training improves generalization of language models to unseen verifiable output constraints.
-
Steered Generation via Gradient-Based Optimization on Sparse Query Features
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
-
MemOS: A Memory OS for AI System
MemOS introduces a unified memory management framework for LLMs using MemCubes to handle and evolve different memory types for improved controllability and evolvability.
-
ReGA: Model-Based Safeguard for LLMs via Representation-Guided Abstraction
ReGA uses safety-critical representations to guide abstraction in model-based analysis, enabling scalable detection of harmful LLM inputs with reported AUROC of 0.975 at prompt level.