Mental health disorders remain among the leading cause of disability worldwide, yet conditions such as depression, anxiety, and Post-Traumatic Stress Disorder (PTSD) are frequently underdiagnosed or misdiagnosed due to subjective assessments, limited clinical resources, and stigma and low awareness. In primary care settings, studies show that providers misidentify depression or anxiety in over 60% of cases, highlighting the urgent need for scalable, accessible, and context-aware diagnostic tools that can support early detection and intervention. In this study, we evaluate the effectiveness of machine learning models for mental health screening using a unique dataset of 553 real-world, semistructured interviews, each paried with ground-truth diagnoses for major depressive episodes (MDE), anxiety disorders, and PTSD. We benchmark multiple model classes, including zero-shot prompting with GPT-4.1 Mini and MetaLLaMA, as well as fine-tuned RoBERTa models using LowRank Adaptation (LoRA). Our models achieve over 80% accuracy across diagnostic categories, with especially strongperformance on PTSD (up to 89% accuracy and 98% recall). We also find that using shorter context, focused context segments improves recall, suggesting that focused narrative cues enhance detection sensitivity. LoRA fine-tuning proves both efficient and effective, with lower-rank configurations (e.g., rank 8 and 16) maintaining competitive performance across evaluation metrics. Our results demonstrate that LLM-based models can offer substantial improvements over traditional self-report screening tools, providing a path toward low-barrier, AI-powerd early diagnosis. This work lays the groundwork for integrating machine learning into real-world clinical workflows, particularly in low-resource or high-stigma environments where access to timely mental health care is most limited.

The fields of explainable AI and mechanistic interpretability aim to uncover the internal structure of neural networks, with circuit discovery as a central tool for understanding model computations. Existing approaches, however, rely on manual inspection and remain limited to toy tasks. Automated interpretability offers scalability by analyzing isolated features and their activations, but it often misses interactions between features and depends strongly on external LLMs and dataset quality. Transcoders have recently made it possible to separate feature attributions into input-dependent and input-invariant components, providing a foundation for more systematic circuit analysis. Building on this, we propose WeightLens and CircuitLens, two complementary methods that go beyond activation-based analysis. WeightLens interprets features directly from their learned weights, removing the need for explainer models or datasets while matching or exceeding the performance of existing methods on context-independent features. CircuitLens captures how feature activations arise from interactions between components, revealing circuit-level dynamics that activation-only approaches cannot identify. Together, these methods increase interpretability robustness and enhance scalable mechanistic analysis of circuits while maintaining efficiency and quality.

Test-time scaling is a powerful strategy for boosting the performance of large language models on complex reasoning tasks. While state-of-the-art approaches often employ generative verifiers to select the best solution from a pool of candidates, this method incurs prohibitive computational costs, limiting its practicality. In this work, we shift the focus to a more budget-aware paradigm: discriminative verification. We conduct a thorough empirical analysis and demonstrate that while discriminative verifiers may underperform in isolation, combining them with self-consistency in a hybrid approach creates a powerful and efficient test-time scaling mechanism. Notably, under a fixed compute budget, this hybrid approach surpasses state-of-the-art generative verification by a significant margin: achieving up to 15.3\% higher accuracy on AIME2025. Our findings establish that for practical, real-world applications, budget-aware scaling with discriminative verifiers is not only a "free" upgrade over self-consistency, but also a more effective and efficient alternative to costly generative techniques. Code is available at https://github.com/wang-research-lab/verification.

Frontier reasoning models have exhibited incredible capabilities across a wide array of disciplines, driven by posttraining large language models (LLMs) with reinforcement learning (RL). However, despite the widespread success of this paradigm, much of the literature has been devoted to disentangling truly novel behaviors that emerge during RL but are not present in the base models. In our work, we approach this question from a different angle, instead asking whether comparable reasoning capabilites can be elicited from base models at inference time by pure sampling, without any additional training. Inspired by Markov chain Monte Carlo (MCMC) techniques for sampling from sharpened distributions, we propose a simple iterative sampling algorithm leveraging the base models' own likelihoods. Over different base models, we show that our algorithm offers substantial boosts in reasoning that nearly match and even outperform those from RL on a wide variety of single-shot tasks, including MA TH500, HumanEval, and GPQA. Moreover, our sampler avoids the collapse in diversity over multiple samples that is characteristic of RL-posttraining. Crucially, our method does not require training, curated datasets, or a verifier, suggesting broad applicability beyond easily verifiable domains.

The Enterprise Intelligence Platform must integrate logs from numerous third-party vendors in order to perform various downstream tasks. However, vendor documentation is often unavailable at test time. It is either misplaced, mismatched, poorly formatted, or incomplete, which makes schema mapping challenging. We introduce a reinforcement learning agent that can self-improve without labeled examples or model weight updates. During inference, the agent: 1) Identifies ambiguous field-mapping attempts. 2) Generates targeted web-search queries to gather external evidence. 3) Applies a confidence-based reward to iteratively refine its mappings. To demonstrate this concept, we converted Microsoft Defender for Endpoint logs into a common schema. Our method increased mapping accuracy from 56.4\%(LLM-only) to 72.73\%(RAG) to 93.94\% over 100 iterations using GPT-4o. At the same time, it reduced the number of low-confidence mappings requiring expert review by 85\%. This new approach provides an evidence-driven, transparent method for solving future industry problems, paving the way for more robust, accountable, scalable, efficient, flexible, adaptable, and collaborative solutions.

Large Language Models (LLMs) are increasingly deployed as autonomous agents, yet their practical utility is fundamentally constrained by a limited context window and state desynchronization resulting from the LLMs' stateless nature and inefficient context management. These limitations lead to unreliable output, unpredictable behavior, and inefficient resource usage, particularly when interacting with large, structured, and sensitive knowledge systems such as codebases and documents. To address these challenges, we introduce the Gatekeeper Protocol, a novel, domain-agnostic framework that governs agent-system interactions. Our protocol mandates that the agent first operate and reason on a minimalist, low-fidelity "latent state" representation of the system to strategically request high-fidelity context on demand. All interactions are mediated through a unified JSON format that serves as a declarative, state-synchronized protocol, ensuring the agent's model of the system remains verifiably grounded in the system's reality. We demonstrate the efficacy of this protocol with Sage, a reference implementation of the Gatekeeper Protocol for software development. Our results show that this approach significantly increases agent reliability, improves computational efficiency by minimizing token consumption, and enables scalable interaction with complex systems, creating a foundational methodology for building more robust, predictable, and grounded AI agents for any structured knowledge domain.

Multimodal large language models (MLLMs) have achieved remarkable performance across diverse vision-and-language tasks. However, their potential in face recognition remains underexplored. In particular, the performance of open-source MLLMs needs to be evaluated and compared with existing face recognition models on standard benchmarks with similar protocol. In this work, we present a systematic benchmark of state-of-the-art MLLMs for face recognition on several face recognition datasets, including LFW, CALFW, CPLFW, CFP, AgeDB and RFW. Experimental results reveal that while MLLMs capture rich semantic cues useful for face-related tasks, they lag behind specialized models in high-precision recognition scenarios in zero-shot applications. This benchmark provides a foundation for advancing MLLM-based face recognition, offering insights for the design of next-generation models with higher accuracy and generalization. The source code of our benchmark is publicly available in the project page.

Recently, many language models have been pretrained with a "midtraining" phase, in which higher quality, often instruction-formatted data, is mixed in at the end of pretraining. Despite the popularity of this practice, there is little scientific understanding of this phase of model training or why it is effective. In this work, we conduct the first systematic investigation of midtraining through controlled experiments with language models pretrained from scratch and fine-tuned on supervised finetuning datasets in different domains. We find that when compared after supervised fine-tuning, the effectiveness of midtraining is highest in the math and code domains, where midtraining can best reduce the syntactic gap between pretraining and posttraining data. In these cases, midtraining consistently outperforms continued pretraining in both in-domain validation loss as well as pretraining data forgetting after posttraining. We conduct ablations on the starting time of the midtraining phase and mixture weights of the midtraining data, using code midtraining as a case study, and find that timing has a greater impact than mixture weights, with earlier introduction of specialized data, yielding greater benefits in-domain as well as preserving general language modeling better. These findings establish midtraining as a domain adaptation technique that compared to continued pretraining yields better performance through reduced forgetting.

Mixture-of-Experts (MoE) models achieve efficient scaling through sparse expert activation, but often suffer from suboptimal routing decisions due to distribution shifts in deployment. While existing test-time adaptation methods could potentially address these issues, they primarily focus on dense models and require access to external data, limiting their practical applicability to MoE architectures. However, we find that, instead of relying on reference data, we can optimize MoE expert selection on-the-fly based only on input context. As such, we propose \textit{a data-free, online test-time framework} that continuously adapts MoE routing decisions during text generation without external supervision or data. Our method cycles between two phases: During the prefill stage, and later in regular intervals, we optimize the routing decisions of the model using self-supervision based on the already generated sequence. Then, we generate text as normal, maintaining the modified router until the next adaption. We implement this through lightweight additive vectors that only update router logits in selected layers, maintaining computational efficiency while preventing over-adaptation. The experimental results show consistent performance gains on challenging reasoning tasks while maintaining robustness to context shifts. For example, our method achieves a 5.5\% improvement on HumanEval with OLMoE. Furthermore, owing to its plug-and-play property, our method naturally complements existing test-time scaling techniques, e.g., achieving 6\% average gains when incorporated with self-consistency on DeepSeek-V2-Lite.

A typical approach developers follow to influence an LLM's behavior in an application is through careful manipulation of the prompt, such as by adding or modifying instructions. However, merely adding more instructions provides little assurance that they will actually be followed. We introduce Instruction Boosting as a post-generation method to increase the reliability of LLM prompt instructions. We show that Instruction Boosting improves the instruction following rate by up to 7 points for two instructions and up to 4 points for ten instructions. To demonstrate these results we introduce SCALEDIF, a benchmark with a scaled instruction volume of up to ten instructions per data sample. We also present an analysis of the commonly observed trend that performance degrades as more instructions are added. We show that an important factor contributing to this trend is the degree of tension and conflict that arises as the number of instructions is increased. We contribute a quantitative conflict scoring tool that explains the observed performance trends and provides feedback to developers on the impact that additional prompt instructions have on a model's performance.