arXiv:2604. 07650v2 Announce Type: replace Abstract: The rapid growth of the large language model (LLM) ecosystem raises a critical question: are seemingly diverse models truly independent?
arXiv:2607. 18114v1 Announce Type: cross Abstract: Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer.
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.
Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer. We study where this susceptibility, spanning sycophancy and related cue-induced biases, lives inside the model.
The paper introduces a reference‑based bias detection method that audits hidden‑state representations of language models by encoding sentences as similarities to a fixed set of anchor sentences. This relative representation allows comparison across model variants, such as before and after fine‑tuning, and yields a metric called Representational Bias Shift (ΔB). ΔB correlates strongly with output‑level bias changes, can detect bias‑increasing checkpoints with high ROC AUC, and is computationally efficient, requiring only a few minutes and far less compute than traditional benchmarks.
The paper proposes a distributional theory explaining how large language models (LLMs) incorporate external evidence into their decision-making process. It identifies three key predictions: (1) evidence is more persuasive when it aligns with the model’s prior beliefs, (2) models more readily accept errors from their own internal processes than from external sources, and (3) the same evidence can improve weaker models while harming stronger ones. Extensive experiments across ten million trials, twelve LLMs from four families, and eight domains—including quantum mechanics, physics, genetics, and molecular biology—confirm these predictions and reveal that evidence integration occurs late in the network as a structured sequence of steps rather than through a simple trust metric.
arXiv:2607. 08065v1 Announce Type: new Abstract: LLM-as-judge (Zheng et al.
arXiv:2609.22219v1 Announce Type: cross Abstract: Evaluating an aligned language model by reading its answers assumes the answers carry the distinction the evaluator cares about. We introduce LLM end...
The paper investigates how large language models (LLMs) used as judges in absolute scoring tasks exhibit systematic biases that compromise reliability. It shows that a judge’s task accuracy strongly predicts both its judging accuracy and its directional bias, yet more capable examinee models consistently receive more lenient judgments. To mitigate these biases, the authors propose a calibrated weighted majority voting (WMV) ensemble that estimates judges’ error rates from inter-judge agreement patterns, achieving near-oracle performance without labeled data and improving both accuracy and fairness.
arXiv:2609.14754v1 Announce Type: cross Abstract: Causal claims about large language model (LLM) internals rest on measurements. Those might include a projection, a cosine, an ablation delta, or an i...
arXiv:2510. 06096v3 Announce Type: replace Abstract: The objectives that Large Language Models (LLMs) implicitly optimize remain dangerously opaque, making trustworthy alignment and auditing a grand challenge.
arXiv:2606. 00005v1 Announce Type: new Abstract: We present the Consilium Protocol, a Byzantine Fault Tolerance-derived architecture for structured multi-model AI deliberation that treats inter-model disagreement as epistemic signal rather than error.
arXiv:2606. 29713v1 Announce Type: cross Abstract: Hallucination is the reliability bottleneck for LLM-based agents, and fact attribution verifiers are the last line of defense -- yet today's verifiers emit only opaque binary labels, leaving agents unable to self-correct and operators unable to audit.