Function vectors steer LLMs successfully where the logit lens fails to decode the target answer, showing the two properties come apart.
hub Canonical reference
In-context vectors: Making in context learning more effective and controllable through latent space steering
Canonical reference. 80% of citing Pith papers cite this work as background.
11
Pith papers citing it
Background
80% of classified citations
hub tools
citation-role summary
background 4
method 1
citation-polarity summary
representative 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.
Translation function vectors extracted from English to one target language improve correct token ranking for translations to multiple other unseen target languages in decoder-only multilingual LLMs.
CODI compresses explicit CoT into continuous space via self-distillation and is the first implicit method to match explicit CoT performance on GSM8k at GPT-2 scale with 3.1x compression and 28.2% higher accuracy than prior implicit approaches.
Activation Addition steers language models by adding contrastive activation vectors from prompt pairs to control high-level properties like sentiment and toxicity at inference time without training.
DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and JailBreakV while preserving general capabilities.
The Master Key Hypothesis states that capabilities are low-dimensional directions transferable across models through linear subspace alignment, with UNLOCK demonstrating gains such as 12.1% accuracy improvement on MATH when transferring CoT from 14B to 7B models.
A new framework using Task Subspace Logit Attribution localizes attention heads specialized for task recognition and task learning in in-context learning, showing they align and rotate hidden states within a task subspace.
Disentangled Safety Adapters decouple safety computations from task-optimized LLMs via lightweight adapters, yielding up to 53% better AUC on safety tasks and dynamic inference-time alignment with reduced performance trade-offs.
Contrastive Activation Addition steers Llama 2 Chat by adding averaged residual-stream activation differences from contrastive example pairs to control targeted behaviors at inference time.
MemOS introduces a unified memory management framework for LLMs using MemCubes to handle and evolve different memory types for improved controllability and evolvability.
citing papers explorer
-
Steerable but Not Decodable: Function Vectors Operate Beyond the Logit Lens
Function vectors steer LLMs successfully where the logit lens fails to decode the target answer, showing the two properties come apart.
-
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.
-
Exploring Language-Agnosticity in Function Vectors: A Case Study in Machine Translation
Translation function vectors extracted from English to one target language improve correct token ranking for translations to multiple other unseen target languages in decoder-only multilingual LLMs.
-
CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation
CODI compresses explicit CoT into continuous space via self-distillation and is the first implicit method to match explicit CoT performance on GSM8k at GPT-2 scale with 3.1x compression and 28.2% higher accuracy than prior implicit approaches.
-
Steering Language Models With Activation Engineering
Activation Addition steers language models by adding contrastive activation vectors from prompt pairs to control high-level properties like sentiment and toxicity at inference time without training.
-
Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs
DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and JailBreakV while preserving general capabilities.
-
The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment
The Master Key Hypothesis states that capabilities are low-dimensional directions transferable across models through linear subspace alignment, with UNLOCK demonstrating gains such as 12.1% accuracy improvement on MATH when transferring CoT from 14B to 7B models.
-
Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis
A new framework using Task Subspace Logit Attribution localizes attention heads specialized for task recognition and task learning in in-context learning, showing they align and rotate hidden states within a task subspace.
-
Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment
Disentangled Safety Adapters decouple safety computations from task-optimized LLMs via lightweight adapters, yielding up to 53% better AUC on safety tasks and dynamic inference-time alignment with reduced performance trade-offs.
-
Steering Llama 2 via Contrastive Activation Addition
Contrastive Activation Addition steers Llama 2 Chat by adding averaged residual-stream activation differences from contrastive example pairs to control targeted behaviors at inference time.
-
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.