🚀 New Paper Alert 🚀 Excited to share that our paper, “Counter with peace! Persona-Guided Discrete Stochastic Processes for Conditioned Counterspeech Generation,” has been accepted to the Findings Track of #WSDM2027! 🎉
👥 @aswiniandpunter, Apoorv Jain, @Tanmoy_Chak

Congratulations to Tamal Majumder, a PhD student at our lab, on getting the #BestPosterAward at the #IITDelhi #OpenHouse2026! 🏆
We are delighted to see his work being recognized and wish him continued success. Congratulations, Tamal! 👏

🚀 #EMNLPFindings2026 #SneakPeek 🚀
What if an AI model sounds confident but fails for entirely different reasons? Can a single confidence score tell us why? Introducing a behavioral diagnostic framework for understanding why multimodal models fail.
📄 aclanthology.org/2026.findings-….

✅ HallusionBench: AUROC rises from 0.634 to 0.769.
✅ VizWiz: AUROC rises from 0.707 to 0.817.
✅ Pooled XGBoost: AUROC improves from 0.78 with scalar confidence to 0.95 with the three diagnostic signals and 0.97 when confidence is added.

🔑 The takeaway is that multimodal uncertainty should diagnose the structure of a failure before deciding whether to abstain or correct. #MultimodalLLMs #Hallucination #AIResearch #UncertaintyEstimation #ResponsibleAI #TrustworthyAI

🚀 #EMNLPFindings2026 #SneakPeek 🚀
How can we generate realistic, psychologically grounded mental health counseling conversations? MAGneT introduces a coordinated multi-agent framework for generating synthetic multi-turn counseling sessions.
📄 aclanthology.org/2026.findings-…

✅ Experts preferred MAGneT-generated sessions in 77.2% of cases on average across nine dimensions.
✅ Generated sessions achieved 3.2% higher general counseling skills and 4.3% higher CBT-specific skills on CTRS than the strongest baseline.

The takeaway: better counseling data requires coordinated psychological techniques and rigorous evaluation not just fluent dialogue.
#NLProc #MultiAgentSystems #MentalHealthAI #SyntheticData #ResponsibleAI

🚀 #EMNLPFindings2026 #SneakPeek 🚀
Can we uncover the hidden limitations of scientific papers that reviews may overlook? Tree-of-Concerns introduces a multi-agent framework for extracting unstated limitations through structured critique.
📜 aclanthology.org/2026.findings-…

We introduce ToC-Bench, a benchmark containing 414 research papers and 1,905 unstated limitations.
Tree-of-Concerns achieves:
✅ 79% higher precision
✅ 11% higher coverage
On the held-out evaluation, it achieves 40.3% precision and 36.1% Coverage@10.

The takeaway: effective scientific critique requires more than multiple AI opinions. It needs diverse perspectives, adversarial scrutiny, and careful reconciliation. #NLProc #MultiAgentSystems #LLMs #PeerReview #AIforScience

📰 Media Feature Alert 📢
Glad to see our perspectives on efficient AI and Small Language Models featured in Lokmat Times! The future of AI is not just about building larger models, but about making them smaller, more efficient, and accessible. @Tanmoy_Chak
#EfficientAI #SLMs