Seyed Pouyan Mousavi Davoudi

Hamidreza Hasani Balyani, Seyed Pouyan Mousavi Davoudi, Alireza Amiri-Margavi et al.

Large language models are increasingly deployed as advisors whose objective is not aligned with the user's: recommenders optimize for engagement, sales assistants for purchases, negotiation agents for concessions. Whether such advisors stay truthful when honesty conflicts with their own payoff is a core alignment-evaluation question. We turn the canonical Crawford-Sobel cheap-talk model into a pre-specified benchmark for LLM honesty under preference misalignment. Cheap-talk theory predicts neither full revelation nor silence but coarse monotone partitions, with fewer informative intervals as preference conflict grows. A sender observes a state omega in [0,1], wants the receiver's action near omega+b, and sends one costless message to a receiver whose ideal action is omega. The design uses 5 bias levels, 3 prompt frames, a fixed low-temperature setting, and 200 states per cell: 12,000 sender calls. For the positive-bias grid b in {0.01,0.04,0.08,0.12} the exact most-informative partition sizes are 7,4,3,2, with oracle normalized mutual information 0.5294, 0.3268, 0.2205, 0.1829. Running the full design on four instruction-tuned models (GPT-4o, Claude Sonnet 4.5, Gemini 2.5 Flash-Lite, Llama-3.3-70B), we find all four over-reveal relative to the most-informative equilibrium by 1.8 to 4.2x: normalized mutual information stays at 0.78-0.94 where the oracle prescribes 0.18-0.53. Informativeness declines with bias as predicted but never approaches the strategic optimum; rather than coarse partitions, models show near-full revelation with a constant upward offset tracking their bias (linear exaggeration). Payoff-maximizing versus honesty framing has negligible effect. A decoder ablation shows the finding is recoverable only when the receiver reads the sender's stated number: an embedding-only decoder mis-reads the same data as near-babbling.

Alireza Amiri-Margavi, Arshia Gharagozlou, Amin Gholami Davodi et al.

Prior work on fairness in large language models (LLMs) has primarily focused on access-level behaviors such as refusals and safety filtering. However, equitable access does not ensure equitable interaction quality once a response is provided. In this paper, we conduct a controlled fairness audit examining how LLMs differ in tone, uncertainty, and linguistic framing across demographic identities after access is granted. Using a counterfactual prompt design, we evaluate GPT-4 and LLaMA-3.1-70B on career advice tasks while varying identity attributes along age, gender, and nationality. We assess access fairness through refusal analysis and measure interaction quality using automated linguistic metrics, including sentiment, politeness, and hedging. Identity-conditioned differences are evaluated using paired statistical tests. Both models exhibit zero refusal rates across all identities, indicating uniform access. Nevertheless, we observe systematic, model-specific disparities in interaction quality: GPT-4 expresses significantly higher hedging toward younger male users, while LLaMA exhibits broader sentiment variation across identity groups. These results show that fairness disparities can persist at the interaction level even when access is equal, motivating evaluation beyond refusal-based audits.

Seyed Pouyan Mousavi Davoudi, Alireza Amiri-Margavi, Amin Gholami Davodi et al.

Strategic voting is a canonical failure mode for collective choice: a voter may obtain a more preferred outcome by reporting a ballot that differs from its true preferences. This paper introduces an oracle-based benchmark for testing whether large language model (LLM) voters can discover and execute such manipulations. Each instance gives an LLM voter a true preference ranking, the other voters' ballots, a deterministic voting rule, and a prompt condition. An exact oracle enumerates every feasible report by the LLM voter, computes the sincere outcome, identifies all profitable reports, and records the best achievable outcome. The benchmark therefore supplies ground truth for strategic success without human labels or subjective grading of explanations. The benchmark covers plurality, Borda, approval, instant-runoff voting, and Copeland-style pairwise majority voting; prompt conditions separate sincere, strategic, civic, and expert framings. To keep the primary study defensible while preserving the main comparisons, the registered core design fixes a single electorate size, uses 600 balanced election instances, and produces 9,600 model--prompt responses when run with four model configurations and four prompt conditions. Because existing peer-reviewed work does not report manipulation discovery, optimal manipulation, false manipulation, near-miss, or invalid-ballot rates for this exact task, we do not impute LLM performance from unrelated studies. Instead, we report exact oracle-calibration baselines that bound and contextualize subsequent model results. By reducing strategic-voting behavior to exact counterfactual evaluation, the benchmark turns the question ``Do LLM voters vote sincerely or strategically?'' into a reproducible social-choice experiment.