The paper investigates how truth representations in small language models are structured. Using a training‑free axis derived from the dominant singular vector of hidden‑state differences between true and false minimal pairs, the authors evaluate 14 models across six architectural families, including Mixture‑of‑Experts. The study examines whether a single direction captures truth, which components contribute, and how this applies to categories with computed truth values.
By Francesco Karim Vicidomini
arXiv:2601. 06599v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) often encode whether a statement is true as a vector in their residual stream activations.
By Shivam Adarsh, Maria Maistro, Christina Lioma
arXiv:2610.00910v1 Announce Type: cross
Abstract: Human reasoning depends on how objects are related within propositions. \textit{How do relations organize the language representations of contextual...
By Yufa Zhou
arXiv:2607. 18305v1 Announce Type: cross Abstract: Some limits on what language models know are not gaps in data coverage but structural properties of learning from text.
By Priyansh Srivastava, Romit Chatterjee
The study investigates how large language models encode moral knowledge by training linear probes for each category of Moral Foundations Theory. It finds that the model’s representations for different moral foundations occupy distinct, largely independent dimensions yet share a common positive component, indicating an integrated but nuanced moral structure. This geometry is consistent across model architectures and scales, emerges early in pre‑training, and reflects corpus statistics rather than the individualizing/binding distinction of the theory.
By Orion Reblitz-Richardson
arXiv:2607. 10248v1 Announce Type: cross Abstract: Language builds discourse contexts other than the actual: a painting, a belief, a memory, a hypothetical.
By Oliver Steele, Jiangtao Wen, Yuxing Han
The paper investigates how the structural entanglement of facts within a large language model’s knowledge base influences whether those facts leak after unlearning. Using two unlearning algorithms (WHP and GA+KL) across fictional and real-world datasets and multiple model sizes, the authors find that highly entangled facts are more likely to be recalled before unlearning, but the relationship changes—WHP weakens it while GA+KL reverses it. By directly manipulating entanglement scores and observing corresponding recall changes, they demonstrate a causal link and develop a predictive tool to audit prompts for potential leakage.
By Aakriti Shah, Yifan Hu, Thai Le
arXiv:2606. 12629v3 Announce Type: replace-cross Abstract: The standard basis of transformer hidden states is a training-free, architecture-general feature basis for detecting concepts and, in language models, steering them; with no learned dictionary.
By Varun Reddy Nalagatla
arXiv:2605.01609v2 Announce Type: replace-cross
Abstract: We find that transformer concept representations systematically anti-concentrate in the spectral tail of the unembedding covariance, encoding...
By Pratyush Acharya, Nuraj Rimal, Habish Dhakal
arXiv:2605. 28149v2 Announce Type: replace Abstract: Sparse Autoencoders (SAEs) extract interpretable features from Large Language Model activations, but standard variants enforce non-negative latents, so a bidirectional semantic axis (e.
By Bartosz Wieciech, Zmnako Awrahman, Marcin Czelej, Victor Hugo Jaramillo Velasquez, Wioletta Stobieniecka
arXiv:2607. 11945v1 Announce Type: cross Abstract: Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world.
By Oliver Steele, Jiangtao Wen, Yuxing Han
The paper investigates how large language models encode moral knowledge by training linear probes for each Moral Foundations Theory category and analyzing their geometric relationships. It finds that the model’s moral directions are largely independent yet share a common component, indicating integration rather than collapse into a single detector. This structure is consistent across architectures, emerges early in pre‑training, and reflects corpus statistics rather than the individualizing/binding distinction of Moral Foundations Theory.