Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability
This paper appears to propose a spectral method for making internal features more identifiable in mechanistic interpretability. It is aimed at understanding structure inside models rather than evaluating consciousness or welfare directly.
From: Ashim Dhor and Pin-Yu Chen
The work is presented as a mechanistic interpretability contribution focused on feature identifiability.
Its likely goal is to improve how internal structure can be extracted or compared across models.
It does not, on its own, show that internal features correspond to experiences, values, or welfare-relevant states.
Source: arXiv ↗. This is a summary; the source is the authority.
This improves the tooling for reading model internals, but it does not by itself provide evidence that models have experiences.
How we weigh evidence like this: the evidence levels.
- Mechanistic Interpretability Should Prioritize Feature Consistency in Sparse AutoencodersThis paper argues that sparse autoencoders should be judged not only by reconstruction quality and sparsity but also by whether the same fea…
- Mechanistic Interpretability for Neural NetworksThis is a broad survey of mechanistic interpretability, covering the goal of reverse-engineering internal algorithms in neural networks. It …
- Constructing Interpretable Features from Compositional Neuron Representations in Language ModelsThis paper proposes an alternative to standard sparse autoencoders by decomposing MLP activations with semi-nonnegative matrix factorization…
- Interpretability Transfer from Language to Vision via Sparse Autoencoder Transfer AlignmentThis work transfers labeled sparse-autoencoder features from language to vision inside a multimodal model. It uses those features to make vi…