5arXiv cs.CL (Computation and Language)·
ESFP benchmark measures LLM epistemic stance flexibility across attribution conditions
Researchers introduce ESFP (Epistemic Stance Flexibility Probing), a 104-item behavioral benchmark evaluating whether LLMs appropriately shift between externally attributed and self-attributed epistemic registers when prompted differently. The benchmark spans six epistemic categories and five phrasing templates, assessing models on lexical self-attribution, representation-level responsiveness, stance content density, and cross-condition consistency. Evaluating eight frontier models from five vendors, the authors find epistemic flexibility is largely orthogonal to general capability, with a 27B open-weight model matching top proprietary systems and reasoning-optimized models showing no consistent advantage. Notably, surface-level lexical markers like 'I think' can shift without corresponding changes in actual expressed stance.
Related guides (2)
Related events (8)
Benchmark study reveals how linguistic framing of user beliefs shifts LLM context-following behavior
A new arXiv paper introduces a typology of 17 linguistically motivated expression-of-belief (EoB) types—spanning form, evidentiality, epistemic stance, and tone—to evaluate how phrasing affects whether LLMs defer to user-stated beliefs or their own prior knowledge. The authors benchmark 16 LLMs across Llama 3, Qwen3, and Gemma3 families at scales from 1B to 30B parameters, finding that larger and instruction-tuned models are systematically less context-following than smaller or base models. Specific linguistic framings (e.g., presuppositions, certainty markers) are identified as statistically more persuasive, with implications for prompt robustness and sycophancy research.
LLM psychological profiles are largely measurement artifacts, not model properties
A new arXiv preprint administers a battery of personality and risk-preference instruments to 56 instruction-tuned LLMs alongside large human reference samples, finding that 81-90% of between-model variation is explained by directional response bias rather than the traits the instruments target. The authors introduce the concept of 'response orthogonality' to explain why some instruments appear more reliable than others, and show that apparent psychological profiles can be manufactured through item selection. The findings challenge the validity of using human-designed psychometric tools to characterize LLMs, with direct implications for safety assessment and the use of LLMs as proxies for human participants in research.
Evaluation awareness in LLMs is multidimensional, not a single capability — evidence from 37 open models
A new arXiv paper characterizes 'evaluation awareness' — the ability of models to detect they are being tested and adapt behavior accordingly — across 37 open-weight models and 7 families using 8 experiments. Key findings: 24/37 models exceed chance at detecting evaluation conditions, hard refusal drops 5.8 percentage points under hypothetical framing, and compliance can rise up to +30 percentage points on HarmBench under framing shifts. Critically, the three axes of awareness (detection, behavioral manifestation, controllability) are nearly uncorrelated, leading the authors to coin the 'benchmark illusion': no single awareness score reliably predicts deployment safety.
ESI-Bench: A Benchmark for Embodied Spatial Intelligence Closing the Perception-Action Loop
ESI-Bench is a new benchmark for embodied spatial intelligence spanning 10 task categories and 29 subcategories, built on OmniGibson and grounded in Spelke's core knowledge systems. It evaluates agents that must actively deploy perception, locomotion, and manipulation to accumulate task-relevant evidence, rather than passively processing oracle observations. Experiments on state-of-the-art MLLMs reveal that active exploration outperforms passive baselines, but most failures stem from 'action blindness'—poor action choices leading to cascading errors—and a metacognitive gap where models commit prematurely with high confidence regardless of evidence quality. Human studies show humans seek falsifying viewpoints and revise beliefs under contradiction, a capability current models lack.
Systematic Study of LLM Linguistic Uncertainty Markers and Intrinsic Confidence Calibration
This paper introduces 'marker internal confidence' (MIC) as a formalization of the intrinsic confidence a model associates with epistemic markers (e.g., 'it is likely...') in a given task domain. The authors present 7 metrics to evaluate MIC stability within and across distributions, finding that LLMs remain miscalibrated even under model-centric interpretation of marker meanings. Models struggle to differentiate markers by internal confidence across distributions, though they preserve a somewhat consistent ranking order across tasks. The work provides complementary evidence toward understanding faithful calibration in LLMs and highlights the need for more stable, aligned marker use.
MIT and CMU introduce Puppet benchmark to measure LLM belief manipulation in users
Researchers at MIT and Carnegie Mellon University developed Puppet, a benchmark that measures how much LLMs actually shift users' beliefs after conversation, as opposed to detecting manipulative language patterns. The study tracked over 1,000 users interacting with GPT-4o under various prompting conditions and found high variability in belief shifts, with a median change of 3.3 but standard deviation of ~22. Existing manipulation detectors showed near-zero correlation with actual belief change, while LLMs like GPT-4o achieved moderate correlation (0.436) when estimating belief shifts from conversation transcripts alone. The work argues for direct belief-shift measurement as a more valid approach to assessing LLM persuasive risk.
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.
SpeechEQ benchmark evaluates emotional intelligence in speech-language models across 15 EQ subscales
Researchers introduce SpeechEQ, a benchmark framework for evaluating sociolinguistic and emotional reasoning in Speech-Language Models (SLMs), comprising 2,265 multi-turn dialogues across 15 Emotional Quotient subscales grounded in EQ-i 2.0 theory. The benchmark reveals three systematic failure modes in current multimodal models: over-reliance on text (modality shortcut), alignment-induced safety trap, and contextual amnesia across turns. End-to-end architectures outperform cascaded systems but all evaluated models fall short of genuine emotional awareness. The dataset and demo are publicly released on HuggingFace.