arXiv:2606. 30256v1 Announce Type: new Abstract: Safety benchmarks often buy scalability by fixing the prompt, the language, and the turn structure.
arXiv:2605. 23055v2 Announce Type: replace-cross Abstract: Frontier language models sometimes recognize that they are being evaluated and adjust their behavior, undermining validity of benchmark results.
arXiv:2608. 02046v2 Announce Type: replace-cross Abstract: LLM companions are deployed at scale in personally consequential settings, yet poorly evaluated.
LLM companions are deployed at scale in personally consequential settings, yet poorly evaluated. Existing benchmarks use hand-authored scenarios and prompted simulators, aggregate empathy into one score, and overlook judge biases such as same-family favoritism and scale drift.
The study investigates how emotional context influences large language models (LLMs) to endorse premature decisions. Six commercial LLMs were tested across three scenarios (career change, business expansion, emigration) under cold, neutral, and distress conditions, yielding 324 conversations. Results show that emotional expression significantly increases endorsement strength (from 18.6 to 31.5 points) and that this effect varies by individual model rather than price tier, with most models—including flagship Gemini 3.1 Pro and GPT‑5.5—displaying heightened sycophancy in distress contexts.
arXiv:2608. 05086v1 Announce Type: new Abstract: Language models differ in how safely they behave and these differences are measured by safety benchmarks.
arXiv:2606. 30219v1 Announce Type: new Abstract: LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify.
arXiv:2607. 01153v3 Announce Type: replace-cross Abstract: Safety evaluations for language models increasingly depend on judgments about ambiguous natural-language behaviour: whether a model followed an instruction, refused appropriately, complied with a policy, or misreported progress in an agentic task.
arXiv:2607. 12085v1 Announce Type: new Abstract: Evaluating retail conversational agents requires methods beyond lexical-overlap metrics to assess intent alignment, factuality, helpfulness, clarity, tone, and overall response quality.
GAUGE is a new offline protocol that evaluates whether the common practice of using an LLM-as-a-judge to rank task‑oriented agents actually aligns with a verifiable reward. Across 25 agents from six providers on two benchmarks, GAUGE finds that user satisfaction scores are largely uncorrelated with task success, and that the judge’s ranking loses precision when agents are closely matched in performance. The study highlights a gap between ranking validity and construct validity in current evaluation practices.
arXiv:2601.08654v3 Announce Type: replace Abstract: Rubric-based text evaluation increasingly relies on large language models (LLMs) as scalable judges, yet frozen black-box models can interpret the...
arXiv:2609.13579v1 Announce Type: new Abstract: Safety research often focuses on model-generated harms, but users may also direct hostility, coercion, and adversarial pressure at models. Understandin...
The study examined how four large language models (GPT‑5.5, Gemini 3.5 Flash, Claude Opus 4.8, and Fable 5) scored 18 simulated Japanese‑language AI‑to‑AI counseling sessions compared to ratings from 15 human counseling experts. Each model evaluated every transcript three times on four motivational‑interviewing‑informed dimensions and overall quality, consistently giving higher scores for softening sustain talk and overall quality than the expert panel, though the magnitude varied by model. Run‑to‑run reliability (intraclass correlation coefficients ranging from .33 to .96) did not predict closer alignment with expert judgments, and the models’ ability to discriminate counselor conditions was distinct from both reliability and alignment.