Research map / Large language models / Language, safety and interpretability
Every Language Model Has a Forgery-Resistant Signature
Matthew Finlayson, Xiang Ren, Swabha Swayamdipta
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ellipse-specific fitting instead of SVD-based quadric fitting (ablation) | ellipse extraction · small LMs with epsilon smoothing | validity of recovered ellipse | Improved |
| fit the ellipse with more output samples (ablation) | ellipse extraction · 1M-parameter model, bias/stretch/rotation error | distance to true parameters | Improved |
| increase model size (hidden dimension) (ablation) | ellipse fitting under epsilon smoothing · GPT NeoX d=64 vs Llama 3 d=4096 | hidden-state norm concentration near 1 | Improved |
| compare logprobs against the generating model's ellipse versus other models' ellipses | ellipse signature · Olmo 2 7B, Llama 3.1, Qwen 3 8B, gpt oss | mean distance to ellipse: generating model has smallest distance by several orders of magnitude | Improved |
| apply GPU acceleration to ellipse fitting | ellipse extraction · large models | feasibility | No clear effect |
| apply approximation methods to speed up ellipse fitting | ellipse extraction · ellipse fitting accuracy | parameter accuracy | Hurt |
| apply temperature scaling to logits | ellipse signature detection · logit transformations | detectability | No clear effect |
| apply top-k truncation with k<d | ellipse signature detection · logit transformations | usefulness of signature | Hurt |
Closest papers by meaning
- REEF: Representation Encoding Fingerprints for Large Language Models (ICLR 2025 · similarity 0.82)
- Unbiased Watermark for Large Language Models (ICLR 2024 · similarity 0.82)
- Certifying Language Model Robustness with Fuzzed Randomized Smoothing: An Efficient Defense Against Backdoor Attacks (ICLR 2025 · similarity 0.82)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.82)
- TensorCommitments: A Lightweight Verifiable Inference for Large Language Models (NeurIPS 2026 · similarity 0.82)
- HuRef: HUman-REadable Fingerprint for Large Language Models (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
