arXiv:2606. 19857v1 Announce Type: cross Abstract: Large language models (LLMs) are commonly prompted and interfaced with human-readable natural language, even when the intended reader is another model.
arXiv:2609.23371v1 Announce Type: cross Abstract: Long-context language models interface with external knowledge through raw natural language. In retrieval-augmented systems, this creates a persisten...
The paper introduces a three-level evaluation framework—behavioral deployment, LM-head readout, and probe recoverability—to distinguish whether a language model fails a syntactic test by not encoding structure or by failing to use it. Using a trilingual control-dependency benchmark, the authors find that probe recoverability consistently exceeds LM-head readout, which in turn exceeds behavioral deployment across seven models and three languages, with the largest gap observed in Qwen3-0.6B Instruct. Layer-localized activation patching shows that instruction tuning shifts the decoded layer later, suggesting decoding favors surface shortcuts and that behavioral evaluation understates what models encode while probing alone overstates what they deploy.
arXiv:2601. 22588v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are widely used as reference-free evaluators via prompting, but this "LLM-as-a-Judge" paradigm is costly, opaque, and sensitive to prompt design.
arXiv:2609.35868v1 Announce Type: new Abstract: Is human readability necessary for effective fine-tuning of large language models? We investigate whether model-conditioned training representations ca...
arXiv:2503. 06211v3 Announce Type: replace-cross Abstract: Text-pretrained language models (LMs) encode rich world knowledge, but adapting them to process and generate perceptual modalities such as audio and images while effectively leveraging that knowledge remains challenging.
The paper argues that large language models cannot achieve perfect reliability for any task, even with unlimited scale. It establishes that each generative task has an inherent reliability ceiling set by how much output uncertainty can be resolved from observable context, with a resolvable part that can be improved by more context and a subjective part tied to task ambiguity. The authors derive a scaling law showing that performance is limited by the scarcer resource—either training data or model capacity—and explain how this law explains phenomena such as retrieval augmentation and catastrophic forgetting.
arXiv:2605. 28969v2 Announce Type: replace-cross Abstract: If an AI agent makes decisions on a person's behalf, those decisions must align with its user.
MechaTerp-TRACE is a new framework that systematically ablates individual components of language models to measure their causal contribution to producing a named entity. By applying TRACE to thirteen instruction‑tuned dense decoder models, the study finds that a small set of positionally fixed components consistently carry the most influence across models and prompts, while the remaining support is evenly distributed. This suggests that entity knowledge is largely embedded in generic generation machinery rather than in isolated, findable components.
The paper introduces Variance‑Calibrated Modulation (VCM), a training‑free pre‑decoding technique that reshapes language model probability distributions before truncation. VCM uses two dynamic mechanisms: a Contextual Searchlight via PMI to suppress stopwords and highlight context‑relevant tokens, and an Adaptive Self‑Debiasing that applies scale‑invariant penalization based on real‑time logit standard deviation. Experiments on open‑ended generation, factual QA, and mathematical reasoning show that VCM consistently reduces the likelihood trap, improving diversity, coherence, and reasoning accuracy with minimal computational cost.
The paper introduces NinaXander, a method for composing frozen language models from different architecture families by inserting a trained shared‑latent adapter between their layers. By running the initial layers of one model, converting the intermediate representation with the adapter, and then continuing with the remaining layers of another model, multiple composed models can be created without retraining. Experiments with RWKV and Pythia show that while some compositions preserve syntactic quality and reduce memory usage, none match the parent model’s accuracy and language‑modeling performance drops on out‑of‑domain data.
arXiv:2607. 14112v1 Announce Type: cross Abstract: Large language models (LLMs) are evaluated as though perfect reliability is achievable for any task given sufficient scale.
arXiv:2607. 03882v1 Announce Type: cross Abstract: LLMs are increasingly deployed as post-hoc explainers of AI-generated outputs, yet it remains unclear whether they can reliably communicate probabilistic information in natural language.