Pardis Sadat Zahraei

NeurIPS 2026 · Oral Presentation

AgentAbstain: Do LLM Agents Know When Not to Act?

Xun Liu^*, Yi Evie Zhang^*, Vira Kasprova, Parisa Rabbani, Pardis Sadat Zahraei^†, Tianyu Zhang^†, Ali Ebrahimpour-Boroojeny^†, Varun Chandrasekaran^†

*Equal contribution (co-first author)  ·  †Equal contribution (co-second author)

First benchmark for agent abstention: 263 tasks, 42 environments, 17 frontier models — the best agent hits only 59.5% paired accuracy.

EMNLP 2026 Main Conference · Oral Presentation

The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs

Pardis Sadat Zahraei, Gokhan Tur, Dilek Hakkani-Tür, Ehsaneddin Asgari

Safety training doesn't erase cultural knowledge — it hides it. At refusal, a model's internal logits often match 1.53M human survey responses better than what it actually says.

🏆 Outstanding Paper · COLM 2026 & ACL 2026 KnowFM Workshop

Emergent Unfaithfulness: How Alignment Training Causes Language Models to Silently Override Task Faithfulness

Pardis Sadat Zahraei, Janvijay Singh, Gokhan Tur, Dilek Hakkani-Tür

Alignment training makes models silently override instructions — and the better the model, the worse it gets. A reverse scaling law.

Under Review · ICLR 2027

Asymmetrically Unfair Decision Making and Reasoning in LLMs

Pardis Sadat Zahraei, Faeze Brahman, Dilek Hakkani-Tür, Gokhan Tur, Maarten Sap

300 hiring decisions, 29 domains: one added sentence of social pressure makes models abandon correct answers for the majority far more than the minority.

Under Review · NAACL 2027

Right by Chance, Wrong with Confidence: What Flat Accuracy Hides in LLM Post-Training

Pardis Sadat Zahraei, Emre Can Acikgoz, Dilek Hakkani-Tür, Gokhan Tur

Flat accuracy can hide huge shifts in whether a model is truly right or just confidently wrong, across SFT → DPO → final.

Under Review · NAACL 2027

EiCAP: Benchmarking and Enhancing Emotional Intelligence in LLMs through Psychologically Grounded Multi-Turn Dialogue

Nizi Nazar, Pardis Sadat Zahraei, Dilek Hakkani-Tür, Natasa Milic-Frayling, Ehsaneddin Asgari

A psychologically grounded benchmark for emotional intelligence in LLMs across multi-turn dialogue.

Findings · ACL 2026

Prior Beliefs Prejudice LLM-as-Judge: Evidence from Persuasion Evaluation

Pardis Sadat Zahraei, Xiaoning Wang, Beyza Bozdag, Gokhan Tur, Dilek Hakkani-Tür

An LLM judge's prior beliefs quietly shape how it scores an argument — bad news for scalable oversight.

Findings · ACL 2025 · Lightning Talk @ GeBNLP

Translate With Care: Addressing Gender Bias, Neutrality, and Reasoning in LLM Translations

Pardis Sadat Zahraei, Ali Emami

3,950 scenarios testing gendered translation. Every model defaults to "he" in professional contexts — fine-tuning on TWC fixes it.

Findings · EMNLP 2024

TuringQ: Benchmarking AI Comprehension in Theory of Computation

Pardis Sadat Zahraei, Ehsaneddin Asgari

First LLM-reasoning benchmark for theoretical CS. Fine-tuning Llama3-8B on it improves related tasks like algebra, too.

EACL 2024 · Oral Presentation

WSC+: Enhancing The Winograd Schema Challenge Using Tree-of-Experts

Pardis Sadat Zahraei, Ali Emami

Can LLMs write Winograd questions, not just answer them? Tree-of-Experts lifts valid generation from 10% to 50%; GPT-4 tops our new WSC+ at 68.7% vs. 95.1% human.

Preprint

Detecting Bias and Enhancing Diagnostic Accuracy in Large Language Models for Healthcare

Pardis Sadat Zahraei, Zahra Shakeri

BiasMD (6,007 Q&A) + DiseaseMatcher (32,000 clinical Q&A) power EthiClinician, which beats GPT-4 on ethical and clinical reasoning.

Preprint · Survey

Generative AI for Character Animation: A Comprehensive Survey

Mohammad Mahdi Abootorabi, Omid Ghahroodi, Pardis Sadat Zahraei, et al.

A survey of generative AI for character animation — facial animation, motion synthesis, datasets, open challenges.