Focusing Condition: Inference-Time Self-Contrastive Steering Elicits Better Conditional Text Embeddings in LLMs

arXiv:2609.32684 · cs.CL · Submitted 2026-09-26 · Read on arXiv

cs.CL

Submitted: 2026-09-26

Updated: 2026-09-26

Code: https://github.com/zifengcheng/SCS

Terminology

• Embeddings — 69× in this paper · explained in 2 episodes
• Steering — 37× in this paper · explained in 2 episodes
• Clustering — 10× in this paper · explained in 2 episodes
• Alignment — 8× in this paper · explained in 1 episode
• Generalization — 8× in this paper · explained in 13 episodes
• Self-Contrastive Steering — 8× in this paper
• Activation Steering — 6× in this paper · explained in 4 episodes
• Contrastive Learning — 6× in this paper · explained in 5 episodes
• Semantic Textual Similarity — 6× in this paper
• Fine-Tuning — 4× in this paper · explained in 5 episodes
• Projection-Based Elimination — 4× in this paper
• Prompt Engineering — 4× in this paper · explained in 4 episodes
• Transformer — 4× in this paper · explained in 2 episodes
• Cosine Similarity — 3× in this paper · explained in 3 episodes
• Direct Elimination — 3× in this paper
• Multi-Head Self-Attention — 3× in this paper
• Computational overhead — 2× in this paper · explained in 2 episodes
• Downstream Tasks — 2× in this paper · explained in 2 episodes
• Generalizability — 2× in this paper · explained in 6 episodes
• Hypernetwork — 2× in this paper · explained in 3 episodes
• Inference-Time Steering — 2× in this paper · explained in 2 episodes
• Accuracy — 1× in this paper · explained in 1 episode
• Activation Addition — 1× in this paper · explained in 1 episode
• Large Language Models (LLMs) — 1× in this paper · explained in 6 episodes
• Representation Engineering — 1× in this paper · explained in 2 episodes
• Semantic Similarity — 1× in this paper · explained in 2 episodes
• Two-stage training — 1× in this paper · explained in 2 episodes

Sources

• LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders
• The Llama 3 Herd of Models
• Towards General Text Embeddings with Multi-stage Contrastive Learning
• WASSA-2017 Shared Task on Emotion Intensity
• Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering
• Text Embeddings by Weakly-Supervised Contrastive Pre-training
• Out-of-the-Box Conditional Text Embeddings from Large Language Models
• Qwen2.5 Technical Report
• Representation Engineering: A Top-Down Approach to AI Transparency

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering