CORTEXA
← Browse
arxivcs.CLcs.LG2026-07-06

Knowledge Knows, Verbalization Tells: Disentangling Latent Directions for Mathematical Solvability in LLMs

Nikolaos Xiros, Maria-Eleni Zoumpoulidi, Georgios Paraskevopoulos

Although LLMs have made significant progress in mathematical reasoning, determining whether a mathematical problem is solvable remains a fundamental yet challenging capability. While recent studies have probed internal representations of model solvability beliefs, verbalization has primarily been studied behaviorally rather than as an internal representation, limiting its analysis and manipulation. We address this gap by separately probing representations of solvability knowledge and verbalization, allowing us to disentangle the two within model hidden states. Across multiple LLMs, we show that knowledge and verbalization are encoded as distinct, linearly decodable representations and that fabrication is primarily associated with changes in verbalization rather than the underlying knowledge. Prompting with unsolvability cues reduces fabrication primarily by shifting verbalization, while activation steering demonstrates that these representations can be echanistically manipulated to improve model abstention.

View free PDFSource page

Related papers

arxivcs.LGcs.AIcs.CLcs.CY2026-07-16

Innocuous-Seeming Data, Latent Ideology: Ideological Generalisation in Finetuned LLMs

Robert Graham, Edward Stevinson, Yariv Barsheshat

Finetuning language models on small, curated datasets is standard practice for adapting them to specific policies or domains. We show that finetuning on narrow, factually-defensible, moderation-passing data can cause broad ideological shifts across unrelated domains, while preser…

View free PDFSource page
arxivcs.CLcs.IRcs.LG2026-07-11

Transforming LLMs into Efficient Cross-Encoders via Knowledge Distillation for RAG Reranking

Shreeya Dasa Lakshminath, Shubhan S

Cross-encoders achieve high reranking accuracy in Retrieval-Augmented Generation (RAG) pipelines but impose quadratic inference costs that limit real-time deployment. We address this by fine-tuning LLaMA 3 (8B) as a drop-in reranker using a two-stage pipeline: supervised fine-tun…

View free PDFSource page
arxivcs.CLcs.AIcs.CYcs.LG2026-07-05

Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5

Lyndon Drake, Zandi Eberstadt

Emergent misalignment (EM) -- the broad misbehaviour a language model acquires after fine-tuning on narrow harmful data -- is mediated in Qwen2.5 models by a latent persona direction, and that direction is causal in open weights. Transplanting it into a model that shares only pre…

View free PDFSource page
arxivcs.CLcs.LG2026-07-21

Scaling Laws for Hypernetwork-Based Knowledge Injection in Large Language Models

Nischay Dhankhar, Dos Baha, Abulhair Saparov

Injecting factual knowledge into large language models (LLMs) reliably and at scale remains an open challenge. Hypernetworks provide a promising solution to large-scale knowledge injection. Although hypernetworks are typically applied for test-time adaptation, we explore their us…

View free PDFSource page