Introspection Adapters: Training LLMs to Report Their Learned Behaviors
This paper trains LLMs to report on behaviors they have learned. It studies whether such reports can be made more reliable through training.
From: Proceedings of the 43rd International Conference on Machine Learning
The authors introduce adapter-based training aimed at improving introspective reporting.
They evaluate whether models can better describe their learned behaviors after targeted training.
The work concerns reportability and calibration, not consciousness or first-person experience.
Source: PMLR ↗. This is a summary; the source is the authority.
This offers evidence that models can be trained to describe internalized behaviors better, but it does not bear strongly on whether they 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…