Recent progress in Multimodal Large Language Models (MLLMs) has highlighted the critical roles of both the visual backbone and the underlying language model. While prior work has primarily focused on scaling these components to billions of parameters, the trade-offs between model size, architecture, and performance remain underexplored. Additionally, inconsistencies in training data and evaluation protocols have hindered direct comparisons, making it difficult to derive optimal design choices. In this paper, we introduce LLaVA-MORE, a new family of MLLMs that integrates recent language models with diverse visual backbones. To ensure fair comparisons, we employ a unified training protocol applied consistently across all architectures. Our analysis systematically explores both small- and medium-scale LLMs -- including Phi-4, LLaMA-3.1, and Gemma-2 -- to evaluate multimodal reasoning, generation, and instruction following, while examining the relationship between model size and performance. Beyond evaluating the LLM impact on final results, we conduct a comprehensive study of various visual encoders, ranging from CLIP-based architectures to alternatives such as DINOv2, SigLIP, and SigLIP2. Additional experiments investigate the effects of increased image resolution and variations in pre-training datasets. Overall, our results provide insights into the design of more effective MLLMs, offering a reproducible evaluation framework that facilitates direct comparisons and can guide future model development. Our source code and trained models are publicly available at: https://github.com/aimagelab/LLaVA-MORE.

AI presents a novel tool for deciphering the motivations behind human behaviors. We show that by varying prompts to a large language model, we can elicit a full range of human behaviors in a variety of different scenarios in terms of classic economic games. Then by analyzing which prompts are needed to elicit which behaviors, we can infer (decipher) the motivations behind the human behaviors. We also show how one can analyze the prompts to reveal relationships between the classic economic games, providing new insight into what different economic scenarios induce people to think about. We also show how this deciphering process can be used to understand differences in the behavioral tendencies of different populations.

Recently, reinforcement learning (RL) has been shown to greatly enhance the reasoning capabilities of large language models (LLMs), and RL-based approaches have been progressively applied to visual multimodal tasks. However, the audio modality has largely been overlooked in these developments. Thus, we conduct a series of RL explorations in audio understanding and reasoning, specifically focusing on the audio question answering (AQA) task. We leverage the group relative policy optimization (GRPO) algorithm to Qwen2-Audio-7B-Instruct, and our experiments demonstrated state-of-the-art performance on the MMAU Test-mini benchmark, achieving an accuracy rate of 64.5%. The main findings in this technical report are as follows: 1) The GRPO algorithm can be effectively applied to large audio language models (LALMs), even when the model has only 8.2B parameters; 2) With only 38k post-training samples, RL significantly outperforms supervised fine-tuning (SFT), indicating that RL-based approaches can be effective without large datasets; 3) The explicit reasoning process has not shown significant benefits for AQA tasks, and how to efficiently utilize deep thinking remains an open question for further research; 4) LALMs still lag far behind humans auditory-language reasoning, suggesting that the RL-based approaches warrant further explorations.

Large language model (LLM) agents need to perform multi-turn interactions in real-world tasks. However, existing multi-turn RL algorithms for optimizing LLM agents fail to perform effective credit assignment over multiple turns while leveraging the generalization capabilities of LLMs and it remains unclear how to develop such algorithms. To study this, we first introduce a new benchmark, ColBench, where an LLM agent interacts with a human collaborator over multiple turns to solve realistic tasks in backend programming and frontend design. Building on this benchmark, we propose a novel RL algorithm, SWEET-RL (RL with Step-WisE Evaluation from Training-time information), that uses a carefully designed optimization objective to train a critic model with access to additional training-time information. The critic provides step-level rewards for improving the policy model. Our experiments demonstrate that SWEET-RL achieves a 6% absolute improvement in success and win rates on ColBench compared to other state-of-the-art multi-turn RL algorithms, enabling Llama-3.1-8B to match or exceed the performance of GPT4-o in realistic collaborative content creation.

Several Department of Commerce bureaus are said to have told staff not to use DeepSeek's AI chatbot on government devices. "To help keep Department of Commerce information systems safe, access to the new Chinese based AI DeepSeek is broadly prohibited on all [government-furnished equipment]," one bureau told staff in an email, Reuters reports. "Do not download, view, access any applications, desktop apps or websites related to DeepSeek." The extent of any DeepSeek ban within the government as a whole is not yet known. Several states -- including New York, Texas and Virginia -- have banned DeepSeek from government devices.

Newer open LLMs often brag about big benchmark improvements, and that was certainly the case with DeepSeek-R1, which came close to OpenAI's o1 in some benchmarks. But the model you run on your Windows laptop isn't the same one that's scoring high marks. This simple question--and the LLM's rambling answer--shows how smaller models can easily go off the rails. They frequently fail to notice context or pick up on nuances that should seem obvious. In fact, recent research suggests that less intelligent large language models with reasoning capabilities are prone to such faults.

The workshop featured 20 accepted papers across diverse application areas, including five oral presentations and 15 posters. Research topics ranged from agent debate evaluation and taxonomy expansion to hypothesis generation, AI4Research benchmarks, caption generation, drug discovery, and financial auditing. Additionally, the workshop hosted a dedicated mentoring session for early-career researchers. The workshop had six inspiring invited talks from academic and industry experts covering a wide range of research topics. Professor Wei Wang (UCLA) presented work on multimodal scientific foundation models for knowledge extraction and synthesis.

There are lots of ways to incorporate AI into robots, starting with improving how they are trained to do tasks. But using large language models to give instructions, as Google has done, is particularly interesting. The robotics startup Figure went viral a year ago for a video in which humans gave instructions to a humanoid on how to put dishes away. Around the same time, a startup spun off from OpenAI, called Covariant, built something similar for robotic arms in warehouses. I saw a demo where you could give the robot instructions via images, text, or video to do things like "move the tennis balls from this bin to that one."

Escape rooms present a unique cognitive challenge that demands exploration-driven planning: players should actively search their environment, continuously update their knowledge based on new discoveries, and connect disparate clues to determine which elements are relevant to their objectives. Motivated by this, we introduce VisEscape, a benchmark of 20 virtual escape rooms specifically designed to evaluate AI models under these challenging conditions, where success depends not only on solving isolated puzzles but also on iteratively constructing and refining spatial-temporal knowledge of a dynamically changing environment. On VisEscape, we observed that even state-of-the-art multimodal models generally fail to escape the rooms, showing considerable variation in their levels of progress and trajectories. To address this issue, we propose VisEscaper, which effectively integrates Memory, Feedback, and ReAct modules, demonstrating significant improvements by performing 3.7 times more effectively and 5.0 times more efficiently on average.

Patient matching is the process of linking patients to appropriate clinical trials by accurately identifying and matching their medical records with trial eligibility criteria. We propose LLM-Match, a novel framework for patient matching leveraging fine-tuned open-source large language models. Our approach consists of four key components. First, a retrieval-augmented generation (RAG) module extracts relevant patient context from a vast pool of electronic health records (EHRs). Second, a prompt generation module constructs input prompts by integrating trial eligibility criteria (both inclusion and exclusion criteria), patient context, and system instructions. Third, a fine-tuning module with a classification head optimizes the model parameters using structured prompts and ground-truth labels. Fourth, an evaluation module assesses the fine-tuned model's performance on the testing datasets. We evaluated LLM-Match on four open datasets - n2c2, SIGIR, TREC 2021, and TREC 2022 - using open-source models, comparing it against TrialGPT, Zero-Shot, and GPT-4-based closed models. LLM-Match outperformed all baselines.