Large language models (LLMs) achieve impressive performance on complex mathematical benchmarks yet sometimes fail on basic math reasoning while generating unnecessarily verbose responses. In this paper, we present LLMThinkBench, a systematic benchmark and comprehensive empirical study to evaluate the efficiency of reasoning in LLMs, focusing on the fundamental tradeoff between accuracy and overthinking. First, we formalize the accuracy-verbosity tradeoff. Second, we introduce the Overthinking Score, a harmonic-mean metric combining accuracy and token-efficiency for holistic model evaluation. Third, we establish an evaluation protocol with dynamically-generated data across 14 basic math tasks. Fourth, we conduct a large-scale empirical study evaluating 53 LLMs, including reasoning and quantized variants across different reasoning budgets. Fifth, we release LLMThinkBench as an open-source Python package and public leaderboard for reproducibility. Our findings reveal: 1) model performance on complex benchmarks does not translate directly to basic math reasoning; 2) reasoning models generate ∼18× more tokens while sometimes achieving lower accuracy and exhibit catastrophic collapse when tokens are constrained, dropping by up to ∼36%; 3) the accuracy-verbosity relationship is non-monotonic with extended reasoning budgets yielding diminishing returns (GPT-5/o-series models show zero accuracy gain from low → medium → high reasoning effort). Our findings challenge the assumption that longer reasoning in LLMs necessarily improves mathematical reasoning. Our public leaderboard is available at https://ctrl-gaurav.github.io/LLMThinkBench/. Our open-source Python package is available at https://pypi.org/project/llmthinkbench/, and the codebase can be found at https://github.com/ctrl-gaurav/LLMThinkBench for easy and reproducible evaluation.

Understanding how the same real-world event is represented across text, video, and audio remains a fundamental challenge in information extraction. While prior datasets have explored events within individual modalities or limited modality pairs, a setting that combines all three modalities remains largely unexplored. We introduce M³EC (Multimodal Multidocument Event Coreference), a new benchmark for multimodal, multidocument event extraction and coreference resolution. It is built from tightly related news clusters that combine eventful videos, audio files, and news articles. We also introduce a semi-automated annotation pipeline, AutoEventAnnotator, to generate automated annotations at scale. Additionally, we construct a human-annotated test set for robust evaluation. To our knowledge, this is the first benchmark to support cluster-level multidocument event extraction and event coreference across text, video, and audio. Evaluation with strong multimodal LLMs shows that the majority of them fail to extract valid events in video and particularly in audio. We further show that models finetuned on the annotations generated by our automated pipeline achieve consistent improvements, including more than +30 F1 on video event extraction, +15 F1 on audio event extraction, and +40 CoNLL F1 on end-to-end coreference resolution, highlighting the usefulness of M³EC for training and evaluation in multimodal event understanding.

Multimodal foundation models that integrate audio, vision, and language achieve strong performance on reasoning and generation tasks, yet their robustness to adversarial manipulation remains poorly understood. We study a realistic and underexplored threat model: untargeted, audio-only adversarial attacks on trimodal audio–video–language models. We analyze six complementary attack objectives that target different stages of multimodal processing, including audio encoder representations, cross-modal attention, hidden states, and output likelihoods. Across four state-of-the-art models and multiple benchmarks, we show that audio-only perturbations can induce severe multimodal failures, achieving up to 96% attack success rate. We further show that attacks can be successful at low perceptual distortions (LPIPS ≤ 0.08, SI-SNR ≥ 0 dB) and benefit more from extended optimization than increased data scale. We evaluate the feasibility of these attacks under physically realistic conditions by incorporating room impulse response (RIR) modeling, showing that audio-only perturbations remain effective under environmental transformations and thus highlight the practical risk of single-modality attacks in real-world multimodal systems. Transferability across models and encoders remains limited, while speech recognition systems such as Whisper primarily respond to perturbation magnitude, achieving >97% attack success under severe distortion. These results expose a previously overlooked single-modality attack surface in multimodal systems and motivate defenses that enforce cross-modal consistency. Our project website is available at https://aafiya-h.github.io/soundbreak/.

Large language models (LLMs) have become increasingly accessible online, thus they can be easily used to generate synthetic data for technology. With the rising capabilities of LLMs, their applications span across many domains. With its increasing use for automating tasks, it is crucial to understand the fairness notions harboured by these models. Our work aims to explore the consistency and behaviour of GPT3.5, GPT-4 in both short-term and long-term scenarios through the lens of fairness. Additionally, the search for an optimal prompt template design for equalized opportunities has been investigated in this study. In the short-term scenario for the German Credit dataset, an intervention to a key feature recorded an increase in loan rejection rate by 37.15% for GPT-3.5 and 49.52% for GPT-4. In the long-term scenario for ML fairness gym, adding extra information about the environment to the prompts has shown no improvement to the prompt with minimal information in terms of final credit distributions. However, adding extra features to the prompt has increased the profit rate by 6.41% (from 17.2% to 23.6%) compared to a baseline maximum-reward classifier with compromising group-level recall rates.

Abstractive Summarization models are generally conditioned on the source article. This would generate a summary with the central theme of the article. However, it would not be possible to generate a summary focusing on specific key areas of the article. To solve this problem, we introduce a novel method for abstractive summarization. We aim to use a transformer to generate summaries which are more tailored to the events in the text by using event information. We extract events from text, perform generalized pooling to get a representation for these events and add an event attention block in the decoder to aid the transformer model in summarization. We carried out experiments on CNN / Daily Mail dataset and the BBC Extreme Summarization dataset. We achieve comparable results on both these datasets, with less training and better inclusion of event information in the summaries as shown by human evaluation scores.