Evaluating Synthetic Activations composed of SAE Latents in GPT-2
원문 초록
Sparse Auto-Encoders (SAEs) are commonly employed in mechanistic interpretability to decompose the residual stream into monosemantic SAE latents. Recent work demonstrates that perturbing a model's activations at an early layer results in a step-function-like change in the model's final layer activations. Furthermore, the model's sensitivity to this perturbation differs between model-generated (real) activations and random activations. In our study, we assess model sensitivity in order to compare real activations to synthetic activations composed of SAE latents. Our findings indicate that synthetic activations closely resemble real activations when we control for the sparsity and cosine similarity of the constituent SAE latents. This suggests that real activations cannot be explained by a simple "bag of SAE latents" lacking internal structure, and instead suggests that SAE latents possess significant geometric and statistical properties. Notably, we observe that our synthetic activations exhibit less pronounced activation plateaus compared to those typically surrounding real activations.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
Generative Latent Kernel Modeling for Blind Motion Deblurring
Differentially Private Relational Learning with Entity-level Privacy Guarantees
EtiCor++: Towards Understanding Etiquettical Bias in LLMs
Fragility in Average Treatment Effect on the Treated under Limited Covariate Support
Assessing parameter identifiability of a hemodynamics PDE model using spectral surrogates and dimension reduction
ACCESS DENIED INC: The First Benchmark Environment for Sensitivity Awareness
토론
이 논문과 연결된 포럼 글이 아직 없습니다.