Vernon Toh

Tej Deep Pala, Vernon Y. H. Toh, Rishabh Bhardwaj et al.

In today's era, where large language models (LLMs) are integrated into numerous real-world applications, ensuring their safety and robustness is crucial for responsible AI usage. Automated red-teaming methods play a key role in this process by generating adversarial attacks to identify and mitigate potential vulnerabilities in these models. However, existing methods often struggle with slow performance, limited categorical diversity, and high resource demands. While Rainbow Teaming, a recent approach, addresses the diversity challenge by framing adversarial prompt generation as a quality-diversity search, it remains slow and requires a large fine-tuned mutator for optimal performance. To overcome these limitations, we propose Ferret, a novel approach that builds upon Rainbow Teaming by generating multiple adversarial prompt mutations per iteration and using a scoring function to rank and select the most effective adversarial prompt. We explore various scoring functions, including reward models, Llama Guard, and LLM-as-a-judge, to rank adversarial mutations based on their potential harm to improve the efficiency of the search for harmful mutations. Our results demonstrate that Ferret, utilizing a reward model as a scoring function, improves the overall attack success rate (ASR) to 95%, which is 46% higher than Rainbow Teaming. Additionally, Ferret reduces the time needed to achieve a 90% ASR by 15.2% compared to the baseline and generates adversarial prompts that are transferable i.e. effective on other LLMs of larger size. Our codes are available at https://github.com/declare-lab/ferret.

Vernon Y. H. Toh, Deepanway Ghosal, Soujanya Poria · deepmind

Large language models (LLMs) have shown increasing competence in solving mathematical reasoning problems. However, many open-source LLMs still struggle with errors in calculation and semantic understanding during intermediate reasoning steps. In this work, we introduce Prove, a simple yet effective framework that leverages translated programs derived from natural language solutions as a verification mechanism to filter out potentially incorrect reasoning paths before aggregating final answers. Unlike vanilla majority voting, our approach filters out solutions whose corresponding program output is inconsistent with the generated solution, aggregating only those that pass verification. We conducted extensive experiments using 13 open-source LLMs from various model families and sizes, ranging from 0.5B to 13B parameters, across eight mathematical benchmarks. Our results show that Prove consistently outperforms vanilla majority voting as a heuristic for solving mathematical reasoning tasks across all model sizes and datasets, achieving improvements of up to 18% on GSM8K and 8% on MATH-500. Our codes are available at https://github.com/declare-lab/prove.

Vernon Y. H. Toh, Yew Ken Chia, Deepanway Ghosal et al. · deepmind

The releases of OpenAI's o-[n] series, such as o1, o3, and o4-mini, mark a significant paradigm shift in Large Language Models towards advanced reasoning capabilities. Notably, models like o3 have demonstrated strong performance on benchmarks like the Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI). However, this benchmark is limited to symbolic patterns, whereas humans often perceive and reason about multimodal scenarios involving both vision and language data. Thus, there is an urgent need to investigate advanced reasoning capabilities in multimodal tasks. To this end, we track the evolution of the GPT-[n] and o-[n] series models (including o1, o3, and o4-mini) on challenging multimodal puzzles from PuzzleVQA and AlgoPuzzleVQA, which demand fine-grained visual perception. Our results reveal that o-[n] series, particularly later iterations like o3 and o4-mini, significantly outperform the GPT-[n] series and show strong scalability in multimodal reasoning. Nonetheless, despite these substantial advancements and the superior capabilities demonstrated by the o-[n] series, our findings highlight that even these leading models face persistent challenges. Difficulties are particularly evident in tasks requiring precise visual perception, robust compositional reasoning across multiple visual attributes, and solving complex algorithmic or highly combinatorial puzzles, indicating critical areas for future AGI development. We plan to continuously track new models in the series and update our results in this paper accordingly. All resources used in this evaluation are openly available at https://github.com/declare-lab/LLM-PuzzleTest.

Qi Sun, Pengfei Hong, Tej Deep Pala et al. · deepmind

Traditional reinforcement learning-based robotic control methods are often task-specific and fail to generalize across diverse environments or unseen objects and instructions. Visual Language Models (VLMs) demonstrate strong scene understanding and planning capabilities but lack the ability to generate actionable policies tailored to specific robotic embodiments. To address this, Visual-Language-Action (VLA) models have emerged, yet they face challenges in long-horizon spatial reasoning and grounded task planning. In this work, we propose the Embodied Multimodal Action Model with Grounded Chain of Thought and Look-ahead Spatial Reasoning, Emma-X. Emma-X leverages our constructed hierarchical embodiment dataset based on BridgeV2, containing 60,000 robot manipulation trajectories auto-annotated with grounded task reasoning and spatial guidance. Additionally, we introduce a trajectory segmentation strategy based on gripper states and motion trajectories, which can help mitigate hallucination in grounding subtask reasoning generation. Experimental results demonstrate that Emma-X achieves superior performance over competitive baselines, particularly in real-world robotic tasks requiring spatial reasoning.

Shang Hong Sim, Tej Deep Pala, Vernon Toh et al.

Generating grounded and trustworthy responses remains a key challenge for large language models (LLMs). While retrieval-augmented generation (RAG) with citation-based grounding holds promise, instruction-tuned models frequently fail even in straightforward scenarios: missing explicitly stated answers, citing incorrectly, or refusing when evidence is available. In this work, we explore how reinforcement learning (RL) and internal reasoning can enhance grounding in LLMs. We use the GRPO (Group Relative Policy Optimization) method to train models using verifiable outcome-based rewards targeting answer correctness, citation sufficiency, and refusal quality, without requiring gold reasoning traces or expensive annotations. Through comprehensive experiments across ASQA, QAMPARI, ELI5, and ExpertQA we show that reasoning-augmented models significantly outperform instruction-only variants, especially in handling unanswerable queries and generating well-cited responses. A two-stage training setup, first optimizing answer and citation behavior and then refusal, further improves grounding by stabilizing the learning signal. Additionally, we revisit instruction tuning via GPT-4 distillation and find that combining it with GRPO enhances performance on long-form, generative QA tasks. Overall, our findings highlight the value of reasoning, stage-wise optimization, and outcome-driven RL for building more verifiable and reliable LLMs.

Brandon Ong, Tej Deep Pala, Vernon Toh et al.

Process Reward Models (PRMs) provide step-level supervision that improves the reliability of reasoning in large language models. While PRMs have been extensively studied in text-based domains, their extension to Vision Language Models (VLMs) remains limited. Existing Vision-Language PRMs (VL-PRMs) rely on Monte Carlo Tree Search (MCTS) for data construction, which can often produce noisy supervision signals and limit generalization across tasks. In this work, we aim to elucidate the design space of VL-PRMs by exploring diverse strategies for dataset construction, training, and test-time scaling. First, we introduce a hybrid data synthesis framework that combines MCTS with judgments from a strong VLM, producing more accurate step-level labels. Second, we propose perception-focused supervision, enabling our PRM to explicitly detect errors at the visual grounding stage of reasoning. Third, we systematically evaluate multiple test-time scaling strategies, showing that our PRMs can reliably guide VLMs toward more accurate solutions. Our experiments covering five diverse multimodal benchmarks (MMMU, PuzzleVQA, AlgoPuzzleVQA, MathVista, and MathVision) reveal several key insights: (i) VL-PRMs when used as Outcome Reward Models (ORMs) during test-time scaling (TTS) can outperform VL-PRM guided process step selection, (ii) smaller VL-PRMs can match or even surpass larger ones in detecting process errors, (iii) VL-PRMs uncover latent reasoning abilities in stronger VLM backbones, (iv) perception-level supervision leads to significant gains in test-time scaling, and (v) TTS performance of different policies improve on advanced math reasoning datasets despite not training VL-PRMs on such datasets. We hope our work will motivate further research and support the advancement of VLMs.