Tag
This paper empirically evaluates three LLM watermarking methods (KGW, Unigram, SynthID-Text) against forensic admissibility standards, finding that none meet the evidentiary bar required by courts, with near-100% removal of watermarks after meaning-preserving paraphrase and high false-negative rates even before attack.
This paper presents the first systematic evaluation of cross-family speculative decoding for Polish LLMs on Apple Silicon, extending MLX-LM with UAG to enable cross-tokenizer decoding. It finds that context-aware token translation improves acceptance rates, but unified memory bandwidth limitations prevent theoretical speedup amortization, with best results showing 1.7x throughput gains for structured text.
This paper presents a comprehensive empirical evaluation of how large language models handle corruptions in chain-of-thought reasoning steps, testing 13 models across 5 perturbation types (MathError, UnitConversion, Sycophancy, SkippedSteps, ExtraSteps) on mathematical reasoning tasks. The findings reveal heterogeneous vulnerability patterns with implications for deploying LLMs in multi-stage reasoning pipelines.