Tool Calling is Linearly Readable and Steerable in Language Models
cs.CL, cs.AI, cs.LG, cs.SE
Submitted: 2026-05-08
Updated: 2026-09-26
Comments: 54 pages. v3: revised manuscript with new experiments
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Refusal in Language Models Is Mediated by a Single Direction
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Not All Language Model Features Are One-Dimensionally Linear
- Gemma: Open Models Based on Gemini Research and Technology
- Dissecting Recall of Factual Associations in Auto-Regressive Language Models
- The Llama 3 Herd of Models
- Internal Representations as Indicators of Hallucinations in Agent Tool Selection
- An Overview of Catastrophic AI Risks
- API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs
- Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
- The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
- Emergent Linear Representations in World Models of Self-Supervised Sequence Models
- The Linear Representation Hypothesis and the Geometry of Large Language Models
- Gorilla: Large Language Model Connected with Massive APIs
- ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs
- Qwen3 Technical Report
- Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
- Linear Representations of Sentiment in Large Language Models
- Function Vectors in Large Language Models
- Steering Language Models With Activation Engineering
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