Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoning

A recent review titled 'LLMs4All' highlights the transformative potential of Large Language Models (LLMs) across various academic disciplines, including arts, economics, and law. The paper emphasizes the capabilities of LLMs, such as ChatGPT, in generating human-like conversations and performing complex language-related tasks, suggesting significant real-world applications in fields like education and scientific discovery.

Recent research has critically evaluated the effectiveness of Reinforcement Learning with Verifiable Rewards (RLVR) in enhancing the reasoning capabilities of large language models (LLMs). The study found that while RLVR-trained models perform better than their base counterparts on certain tasks, they do not exhibit fundamentally new reasoning patterns, particularly at larger evaluation metrics like pass@k.

A new framework called Mujica-MyGo has been proposed to enhance multi-agent Retrieval-Augmented Generation (RAG) systems, addressing the challenges of long context lengths in large language models (LLMs). This framework aims to improve multi-turn reasoning by utilizing a divide-and-conquer approach, which helps manage the complexity of interactions with search engines during complex reasoning tasks.

A recent study on Large Language Models (LLMs) highlights the challenge of context drift in multi-turn interactions, where a model's outputs may diverge from user goals over time. The research introduces a dynamical framework to analyze this drift, formalizing it through KL divergence and proposing a recurrence model to interpret its evolution. This approach aims to enhance the consistency of LLM responses across multiple conversational turns.

ExPO-HM (Explain-then-Detect Policy Optimization for Hateful Memes) has been proposed to enhance the detection of hateful memes, addressing limitations in existing models that primarily provide binary predictions without context. This new approach aims to incorporate reasoning similar to human annotators, improving the understanding of policy-relevant cues such as targets and attack types.

A new framework called ReVeL (Rewrite and Verify by LLM) has been proposed to enhance the multiple-choice question answering (MCQA) format used in evaluating multimodal language models. This framework transforms MCQA into open-form questions while ensuring answers remain verifiable, addressing issues of answer guessing and unreliable accuracy metrics during reinforcement fine-tuning (RFT).

A recent study evaluated the mathematical reasoning capabilities of Large Language Models (LLMs) using the 2026 Korean College Scholastic Ability Test (CSAT) Mathematics section, ensuring a contamination-free evaluation environment. The research involved digitizing all 46 questions immediately after the exam's public release, allowing for a rigorous assessment of 24 state-of-the-art LLMs across various input modalities and languages.

LexInstructEval has been introduced as a new benchmark and evaluation framework aimed at enhancing the ability of Large Language Models (LLMs) to follow complex lexical instructions. This framework utilizes a formal, rule-based grammar to break down intricate instructions into manageable components, facilitating a more systematic evaluation process.