FineWeb-CLaR is introduced, a large-scale annotated dataset derived from FineWeb and FineWeb-2 that places web documents on a shared culture-language-region axis for corpus auditing and benchmark alignment, enabling direct comparison between corpus-side pretraining evidence and benchmark-side evaluation coverage.
Yusser Al Ghussin, Eva Gavaller, Cristina España-Bonet et al.· 0 citations
Chain-of-thought (CoT) can sound plausible yet be unfaithful to the model's underlying reasoning. Most prior work probes CoT faithfulness through input--output behavior or input attributions, leaving internal computation largely underexplored. We instead cast faithfulness as internal concept grounding: Does a large lan...
Qianli Wang, Yilong Wang, Dennis Wei et al.· 0 citations
By structurally decoupling abstract judging from output formatting, this work provides a mechanistic account of format-induced inconsistency on the open-weight models the authors study and implies that benchmark comparisons of judge reliability across formats partly measure the fragile formatting stage, and can underst...
Nils Feldhus, Tanja Baeumel, Elena Golimblevskaia et al.· 1 citation
Many languages are written in multiple scripts, requiring large language models (LLMs) to generate equivalent linguistic content in distinct orthographic forms. While prior work suggests that LLMs route information through shared latent representations, how they internally mediate script variation remains poorly unders...
Daniil Gurgurov, Alan Saji, Katharina Trinley et al.· arXiv.org· 0 citations
This study examines the compositionality of steering vectors for language and behavioral control in large language models. Focusing on language, jailbreak, and conciseness, we investigate whether additive, training-free composition of attribute steering vectors can preserve the intended steering effect of each attribut...
Hyunku Kang, Daniil Gurgurov, Tanja Baeumel et al.· 0 citations
This work qualitatively replicate and extend ConSim's ranking of explanation methods across the tested datasets, explanation families, and simulator LLMs, and identifies two limitations: when class names are meaningful, simulators can obtain high simulatability by solving the classification task directly, without relyi...
Antonin Poché, Fanny Jourdan, Nils Feldhus et al.· 0 citations
It is argued that framing tokenization as output supervision provides a principled account of why tokenization consistently affects model performance, and that differences in task performance, training dynamics, and model internals are induced by output tokenization and largely invariant to input tokenization.
Tanja Baeumel, Josef van Genabith, Simon Ostermann· 0 citations
It is shown that translation is even more modular than previously assumed and that the output language production in translation processes is actually further separable into a syntax and a surface language process.
M. Sonkin, Tanja Baeumel, Daniil Gurgurov et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
