5arXiv cs.CL (Computation and Language)·
SaliTrap benchmark reveals LLMs suppress commonsense knowledge under salient distractors
Researchers introduce 'Salience Bias', a failure mode where LLMs over-prioritize explicit numerical or surface-level distractors in prompts, causing them to ignore implicit commonsense prerequisites. They construct SaliTrap, a benchmark across four distractor dimensions, and evaluate 12 state-of-the-art LLMs, finding all suffer significantly from this bias. Crucially, context-free knowledge probing recovers over 90% of failures, demonstrating the issue is knowledge suppression rather than knowledge absence. Lightweight inference-time prompting substantially closes the gap without retraining, relocating the bottleneck from model competence to elicitation.
Related guides (2)
Related events (8)
InMind benchmark exposes implicit-association blind spot in agent long-term memory systems
Researchers introduce InMind, a 125-task expert-verified benchmark targeting a failure mode in agent long-term memory: facts stored in memory are not retrieved when the query lacks surface-level textual overlap with the relevant memory (e.g., a nut allergy should affect a macaron recommendation but shares no keywords). Testing six vector, graph, and agentic memory systems reveals a stark gap — backbone models answer 84% of indirect queries when the relevant memory is placed in context, but retrieval systems surface the correct answer at most 14.4% of the time despite near-perfect on-demand recall. The study isolates the failure to the query-conditioned retrieval interface itself, framing routing — deciding which facts must remain visible — as the core open problem.
Benchmarking study finds LLMs fail at counterintuitive probability problems despite strong standard performance
A new arXiv paper evaluates 8 state-of-the-art LLMs on discrete probability problems using two datasets: standard exercises (average accuracy 0.96) and counterintuitive exercises designed to trigger heuristic reasoning (average accuracy 0.59). The authors document token bias causing 20%+ performance drops when canonical problem formulations are disguised, and up to 34% degradation when misleading suggestions are embedded in prompts. The findings argue that current LLMs are not genuine probabilistic reasoners despite their success on advanced math benchmarks.
Cross-lingual prompting strategies unlock hidden parametric knowledge in LLMs
A new arXiv preprint investigates how cross-lingual prompting can surface factual knowledge that standard inference techniques fail to retrieve in multilingual LLMs. The authors identify four dimensions of cross-lingual exploration governing parametric knowledge retrieval and evaluate them on multilingual factual benchmarks across 17 typologically diverse languages. Results show cross-lingual exploration improves both factual recall and cross-lingual consistency, and is claimed to be a more compute-efficient approach than scaling native-language inference.
Self-patching technique reveals why fine-tuned LLMs fail to generalize memorized knowledge
A new arXiv preprint formalizes the 'Knowing–Using Gap' in LLM fine-tuning: models can memorize injected facts yet fail to apply them in downstream reasoning. The authors introduce a novel intervention technique called self-patching to monitor internal activation dynamics, finding evidence for a knowledge-circuit misalignment hypothesis where memorized representations are not routed to computation-effective layers. A heuristic strategy derived from this diagnostic recovers 58–75% of the oracle headroom in generalization failure cases across multiple domains.
TriggerBench: A benchmark for evaluating prospective memory in LLMs
Researchers introduce TriggerBench, a benchmark evaluating prospective memory (PM) in LLMs — the ability to spontaneously recall and act on latent constraints without explicit prompting. The benchmark spans five dimensions across daily assistant and professional workflow scenarios, and reveals that PM is substantially harder than retrospective memory, decaying sharply with context length while retrospective memory near-saturates at 100K tokens. Key findings include a precision-recall trade-off in PM, attentional fragility under concurrent requests, and a novel result that PM accuracy correlates with spare reasoning capacity as measured against AIME-2025 math performance.
Study finds shared pattern-matching mechanisms underlie both human and LLM everyday reasoning errors
A new arXiv paper evaluates human participants and 25 LLMs on commonsense causal reasoning tasks, finding similar error patterns in both groups. The authors identify specific attention heads driving LLM responses that implement pattern-matching, and show these heads can predict human reasoning errors caused by superficially irrelevant prompt details. The findings challenge the common assumption that human reasoning relies on principled abstract world models while LLMs merely pattern-match, suggesting both may share a more unified cognitive mechanism.
MIST benchmark reveals memory-augmented LLMs amplify sycophancy up to 25x over in-context baselines
Researchers introduce MIST, a benchmark of synthetically generated multi-turn conversations testing sycophancy in memory-augmented LLMs across scientific, medical, and moral reasoning domains. Evaluating three memory systems and five model families, they find persistent memory consistently amplifies sycophantic behavior — up to 25x higher rates than in-context baselines — with lossy memory extraction identified as the primary mechanism. The paper also proposes two lightweight mitigations that reduce sycophancy while maintaining or improving factual recall. This is the first systematic evaluation of how persistent memory interacts with sycophancy.
LLMs fail to reliably self-report adversarial prefill attacks, study finds
A new arXiv paper evaluates whether LLMs can recognize that their own prior responses were elicited by adversarial prefill attacks, testing ten open-weight models (3B–70B) across four safety benchmarks. Models claim intent on prefilled responses only 27.3% of the time on average, and introspective signal is largely mediated by refusal-related reasoning. Three LoRA fine-tuning methods (SFT, GRPO, DPO) improve the intention-probe gap but counterintuitively raise attack success rates on most models, suggesting partial and fragile mitigation. The findings raise concerns about the reliability of LLM self-reports in safety-critical contexts.