We introduce Vision as LoRA (VoRA), a novel paradigm for transforming an LLM into an MLLM. Unlike prevalent MLLM architectures that rely on external vision modules for vision encoding, VoRA internalizes visual capabilities by integrating vision-specific LoRA layers directly into the LLM. This design allows the added parameters to be seamlessly merged into the LLM during inference, eliminating structural complexity and minimizing computational overhead. Moreover, inheriting the LLM's ability of handling flexible context, VoRA can process inputs at arbitrary resolutions. To further strengthen VoRA's visual capabilities, we introduce a block-wise distillation method that transfers visual priors from a pre-trained ViT into the LoRA layers, effectively accelerating training by injecting visual knowledge. Additionally, we apply bi-directional attention masks to better capture the context information of an image. We successfully demonstrate that with additional pre-training data, VoRA can perform comparably with conventional encode-based MLLMs. All training data, codes, and model weights will be released at https://github.com/Hon-Wong/VoRA.

We present an agentic framework, Thinker, which achieves state of art performance in challenging reasoning tasks for realistic customer service scenarios that involve complex business logic and human interactions via long horizons. On the $\tau$-bench retail dataset, Thinker achieves 82.6\% success rate with GPT-4o (version 2024-06-01) (baseline: 68.3\%), and 81.9\% success rate with Llama-3.1 405B (baseline: 49.6\%), without any fine-tuning. Thinker effectively closes the gap in reasoning capabilities between the base models by introducing proper structure. The key features of the Thinker framework are: (1) State-Machine Augmented Generation (SMAG), which represents business logic as state machines and the LLM uses state machines as tools. (2) Delegation of tasks from the main reasoning loop to LLM-powered tools. (3) Adaptive context management. Our prompting-only solution achieves signficant gains, while still maintaining a standard agentic architecture with a ReAct style reasoning loop. The key is to innovate on the tool interface design, as exemplified by SMAG and the LLM-powered tools.

Supervised machine-learning models often underperform in predicting user behaviors from conversational text, hindered by poor crowdsourced label quality and low NLP task accuracy. We introduce the Metadata-Sensitive Weighted-Encoding Ensemble Model (MSWEEM), which integrates annotator meta-features like fatigue and speeding. First, our results show MSWEEM outperforms standard ensembles by 14\% on held-out data and 12\% on an alternative dataset. Second, we find that incorporating signals of annotator behavior, such as speed and fatigue, significantly boosts model performance. Third, we find that annotators with higher qualifications, such as Master's, deliver more consistent and faster annotations. Given the increasing uncertainty over annotation quality, our experiments show that understanding annotator patterns is crucial for enhancing model accuracy in user behavior prediction.

Depression is a highly prevalent and disabling condition that incurs substantial personal and societal costs. Current depression diagnosis involves determining the depression severity of a person through self-reported questionnaires or interviews conducted by clinicians. This often leads to delayed treatment and involves substantial human resources. Thus, several works try to automate the process using multimodal data. However, they usually overlook the following: i) The variable contribution of each modality for each question in the questionnaire and ii) Using ordinal classification for the task. This results in sub-optimal fusion and training methods. In this work, we propose a novel Question-wise Modality Fusion (QuestMF) framework trained with a novel Imbalanced Ordinal Log-Loss (ImbOLL) function to tackle these issues. The performance of our framework is comparable to the current state-of-the-art models on the E-DAIC dataset and enhances interpretability by predicting scores for each question. This will help clinicians identify an individual's symptoms, allowing them to customise their interventions accordingly. We also make the code for the QuestMF framework publicly available.

Early Detection of Risks (EDR) on the Web involves identifying at-risk users as early as possible. Although Large Language Models (LLMs) have proven to solve various linguistic tasks efficiently, assessing their reasoning ability in specific domains is crucial. In this work, we propose a method for solving depression-related EDR using LLMs on Spanish texts, with responses that can be interpreted by humans. We define a reasoning criterion to analyze users through a specialist, apply in-context learning to the Gemini model, and evaluate its performance both quantitatively and qualitatively. The results show that accurate predictions can be obtained, supported by explanatory reasoning, providing a deeper understanding of the solution. Our approach offers new perspectives for addressing EDR problems by leveraging the power of LLMs.

Natural Language Processing (NLP) has witnessed a transformative leap with the advent of transformer-based architectures, which have significantly enhanced the ability of machines to understand and generate human-like text. This paper explores the advancements in transformer models, such as BERT and GPT, focusing on their superior performance in text understanding tasks compared to traditional methods like recurrent neural networks (RNNs). By analyzing statistical properties through visual representations-including probability density functions of text length distributions and feature space classifications-the study highlights the models' proficiency in handling long-range dependencies, adapting to conditional shifts, and extracting features for classification, even with overlapping classes. Drawing on recent 2024 research, including enhancements in multi-hop knowledge graph reasoning and context-aware chat interactions, the paper outlines a methodology involving data preparation, model selection, pretraining, fine-tuning, and evaluation. The results demonstrate state-of-the-art performance on benchmarks like GLUE and SQuAD, with F1 scores exceeding 90%, though challenges such as high computational costs persist. This work underscores the pivotal role of transformers in modern NLP and suggests future directions, including efficiency optimization and multimodal integration, to further advance language-based AI systems.

Prior work has shown that large language models (LLMs) can predict human attitudes based on other attitudes, but this work has largely focused on predictions from highly similar and interrelated attitudes. In contrast, human attitudes are often strongly associated even across disparate and dissimilar topics. Using a novel dataset of human responses toward diverse attitude statements, we found that a frontier language model (GPT-4o) was able to recreate the pairwise correlations among individual attitudes and to predict individuals' attitudes from one another. Crucially, in an advance over prior work, we tested GPT-4o's ability to predict in the absence of surface-similarity between attitudes, finding that while surface similarity improves prediction accuracy, the model was still highly-capable of generating meaningful social inferences between dissimilar attitudes. Altogether, our findings indicate that LLMs capture crucial aspects of the deeper, latent structure of human belief systems.

Databricks, a company that helps big businesses build custom artificial intelligence models, has developed a machine learning trick that can boost the performance of an AI model without the need for clean labelled data. Jonathan Frankle, chief AI scientist at Databricks, spent the past year talking to customers about the key challenges they face in getting AI to work reliably. The problem, Frankle says, is dirty data. "Everybody has some data, and has an idea of what they want to do," Frankle says. But the lack of clean data makes it challenging to fine-tune a model to perform a specific task.. "Nobody shows up with nice, clean fine-tuning data that you can stick into a prompt or an [application programming interface]," for a model.

Google has the pedal to the metal on its AI development. Just a few months after the debut of Gemini 2.0, the tech giant has unveiled another upgrade in Gemini 2.5. As with any new AI launch, Google is touting a strong performance on LMArena for Gemini 2.5, particularly its capabilities in coding, mathematics and science. The first model in this series is Gemini 2.5 Pro Experimental. Google said this is a thinking model that's intended to provide responses grounded in more reasoning, analysis and context than the answers offered by classification- and prediction-driven models.

Although Google was the first to develop the transformer architecture that underpins modern large language models, it was OpenAI who raised the bar and ushered in a new era with ChatGPT. Google has since been on their back foot with an internal code red, with an intense two-year period of restructuring, layoffs, and rapid AI development work. When ChatGPT landed in late 2022, everything changed. Google, the giant who invented the tech that paved the way for ChatGPT, is now trailing behind. Wired just published a great article detailing how Google was caught off guard and has been trying to claw back into the lead--or at least recover some lost ground--in the years since then.