Sushant Prakash

Xinyun Chen, Renat Aksitov, Uri Alon et al. · deepmind

Self-consistency with chain-of-thought prompting (CoT) has demonstrated remarkable performance gains on various challenging tasks, by utilizing multiple reasoning paths sampled from large language models (LLMs). However, self-consistency relies on the answer extraction process to aggregate multiple solutions, which is not applicable to free-form answers. In this work, we propose Universal Self-Consistency (USC), which leverages LLMs themselves to select the most consistent answer among multiple candidates. We evaluate USC on a variety of benchmarks, including mathematical reasoning, code generation, long-context summarization, and open-ended question answering. On open-ended generation tasks where the original self-consistency method is not applicable, USC effectively utilizes multiple samples and improves the performance. For mathematical reasoning, USC matches the standard self-consistency performance without requiring the answer formats to be similar. Finally, without access to execution results, USC also matches the execution-based voting performance on code generation.

Tao Tu, Shekoofeh Azizi, Danny Driess et al.

Medicine is inherently multimodal, with rich data modalities spanning text, imaging, genomics, and more. Generalist biomedical artificial intelligence (AI) systems that flexibly encode, integrate, and interpret this data at scale can potentially enable impactful applications ranging from scientific discovery to care delivery. To enable the development of these models, we first curate MultiMedBench, a new multimodal biomedical benchmark. MultiMedBench encompasses 14 diverse tasks such as medical question answering, mammography and dermatology image interpretation, radiology report generation and summarization, and genomic variant calling. We then introduce Med-PaLM Multimodal (Med-PaLM M), our proof of concept for a generalist biomedical AI system. Med-PaLM M is a large multimodal generative model that flexibly encodes and interprets biomedical data including clinical language, imaging, and genomics with the same set of model weights. Med-PaLM M reaches performance competitive with or exceeding the state of the art on all MultiMedBench tasks, often surpassing specialist models by a wide margin. We also report examples of zero-shot generalization to novel medical concepts and tasks, positive transfer learning across tasks, and emergent zero-shot medical reasoning. To further probe the capabilities and limitations of Med-PaLM M, we conduct a radiologist evaluation of model-generated (and human) chest X-ray reports and observe encouraging performance across model scales. In a side-by-side ranking on 246 retrospective chest X-rays, clinicians express a pairwise preference for Med-PaLM M reports over those produced by radiologists in up to 40.50% of cases, suggesting potential clinical utility. While considerable work is needed to validate these models in real-world use cases, our results represent a milestone towards the development of generalist biomedical AI systems.

Daniel McDuff, Mike Schaekermann, Tao Tu et al.

An accurate differential diagnosis (DDx) is a cornerstone of medical care, often reached through an iterative process of interpretation that combines clinical history, physical examination, investigations and procedures. Interactive interfaces powered by Large Language Models (LLMs) present new opportunities to both assist and automate aspects of this process. In this study, we introduce an LLM optimized for diagnostic reasoning, and evaluate its ability to generate a DDx alone or as an aid to clinicians. 20 clinicians evaluated 302 challenging, real-world medical cases sourced from the New England Journal of Medicine (NEJM) case reports. Each case report was read by two clinicians, who were randomized to one of two assistive conditions: either assistance from search engines and standard medical resources, or LLM assistance in addition to these tools. All clinicians provided a baseline, unassisted DDx prior to using the respective assistive tools. Our LLM for DDx exhibited standalone performance that exceeded that of unassisted clinicians (top-10 accuracy 59.1% vs 33.6%, [p = 0.04]). Comparing the two assisted study arms, the DDx quality score was higher for clinicians assisted by our LLM (top-10 accuracy 51.7%) compared to clinicians without its assistance (36.1%) (McNemar's Test: 45.7, p < 0.01) and clinicians with search (44.4%) (4.75, p = 0.03). Further, clinicians assisted by our LLM arrived at more comprehensive differential lists than those without its assistance. Our study suggests that our LLM for DDx has potential to improve clinicians' diagnostic reasoning and accuracy in challenging cases, meriting further real-world evaluation for its ability to empower physicians and widen patients' access to specialist-level expertise.

Aileen Cheng, Alon Jacovi, Amir Globerson et al.

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models' world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model's overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts .

Jiaxing Wu, Lin Ning, Luyang Liu et al.

LLM-powered personalization agent systems employ Large Language Models (LLMs) to predict users' behavior from their past activities. However, their effectiveness often hinges on the ability to effectively leverage extensive, long user historical data due to its inherent noise and length of such data. Existing pretrained LLMs may generate summaries that are concise but lack the necessary context for downstream tasks, hindering their utility in personalization systems. To address these challenges, we introduce Reinforcement Learning from Prediction Feedback (RLPF). RLPF fine-tunes LLMs to generate concise, human-readable user summaries that are optimized for downstream task performance. By maximizing the usefulness of the generated summaries, RLPF effectively distills extensive user history data while preserving essential information for downstream tasks. Our empirical evaluation demonstrates significant improvements in both extrinsic downstream task utility and intrinsic summary quality, surpassing baseline methods by up to 22% on downstream task performance and achieving an up to 84.59% win rate on Factuality, Abstractiveness, and Readability. RLPF also achieves a remarkable 74% reduction in context length while improving performance on 16 out of 19 unseen tasks and/or datasets, showcasing its generalizability. This approach offers a promising solution for enhancing LLM personalization by effectively transforming long, noisy user histories into informative and human-readable representations.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Ulrich Rückert, Srinivas Sunkara, Abhinav Rastogi et al.

In human-human conversations, Context Tracking deals with identifying important entities and keeping track of their properties and relationships. This is a challenging problem that encompasses several subtasks such as slot tagging, coreference resolution, resolving plural mentions and entity linking. We approach this problem as an end-to-end modeling task where the conversational context is represented by an entity repository containing the entity references mentioned so far, their properties and the relationships between them. The repository is updated turn-by-turn, thus making training and inference computationally efficient even for long conversations. This paper lays the groundwork for an investigation of this framework in two ways. First, we release Contrack, a large scale human-human conversation corpus for context tracking with people and location annotations. It contains over 7000 conversations with an average of 11.8 turns, 5.8 entities and 15.2 references per conversation. Second, we open-source a neural network architecture for context tracking. Finally we compare this network to state-of-the-art approaches for the subtasks it subsumes and report results on the involved tradeoffs.

Lin Ning, Karan Singhal, Ellie X. Zhou et al.

Deep retrieval models are widely used for learning entity representations and recommendations. Federated learning provides a privacy-preserving way to train these models without requiring centralization of user data. However, federated deep retrieval models usually perform much worse than their centralized counterparts due to non-IID (independent and identically distributed) training data on clients, an intrinsic property of federated learning that limits negatives available for training. We demonstrate that this issue is distinct from the commonly studied client drift problem. This work proposes batch-insensitive losses as a way to alleviate the non-IID negatives issue for federated movie recommendations. We explore a variety of techniques and identify that batch-insensitive losses can effectively improve the performance of federated deep retrieval models, increasing the relative recall of the federated model by up to 93.15% and reducing the relative gap in recall between it and a centralized model from 27.22% - 43.14% to 0.53% - 2.42%. We also open-source our code framework to accelerate further research and applications of federated deep retrieval models.

Karan Singhal, Hakim Sidahmed, Zachary Garrett et al.

Personalization methods in federated learning aim to balance the benefits of federated and local training for data availability, communication cost, and robustness to client heterogeneity. Approaches that require clients to communicate all model parameters can be undesirable due to privacy and communication constraints. Other approaches require always-available or stateful clients, impractical in large-scale cross-device settings. We introduce Federated Reconstruction, the first model-agnostic framework for partially local federated learning suitable for training and inference at scale. We motivate the framework via a connection to model-agnostic meta learning, empirically demonstrate its performance over existing approaches for collaborative filtering and next word prediction, and release an open-source library for evaluating approaches in this setting. We also describe the successful deployment of this approach at scale for federated collaborative filtering in a mobile keyboard application.

Renat Aksitov, Sobhan Miryoosefi, Zonglin Li et al. · deepmind

Answering complex natural language questions often necessitates multi-step reasoning and integrating external information. Several systems have combined knowledge retrieval with a large language model (LLM) to answer such questions. These systems, however, suffer from various failure cases, and we cannot directly train them end-to-end to fix such failures, as interaction with external knowledge is non-differentiable. To address these deficiencies, we define a ReAct-style LLM agent with the ability to reason and act upon external knowledge. We further refine the agent through a ReST-like method that iteratively trains on previous trajectories, employing growing-batch reinforcement learning with AI feedback for continuous self-improvement and self-distillation. Starting from a prompted large model and after just two iterations of the algorithm, we can produce a fine-tuned small model that achieves comparable performance on challenging compositional question-answering benchmarks with two orders of magnitude fewer parameters.

Stephen R. Pfohl, Heather Cole-Lewis, Rory Sayres et al.

Large language models (LLMs) hold promise to serve complex health information needs but also have the potential to introduce harm and exacerbate health disparities. Reliably evaluating equity-related model failures is a critical step toward developing systems that promote health equity. We present resources and methodologies for surfacing biases with potential to precipitate equity-related harms in long-form, LLM-generated answers to medical questions and conduct a large-scale empirical case study with the Med-PaLM 2 LLM. Our contributions include a multifactorial framework for human assessment of LLM-generated answers for biases, and EquityMedQA, a collection of seven datasets enriched for adversarial queries. Both our human assessment framework and dataset design process are grounded in an iterative participatory approach and review of Med-PaLM 2 answers. Through our empirical study, we find that our approach surfaces biases that may be missed via narrower evaluation approaches. Our experience underscores the importance of using diverse assessment methodologies and involving raters of varying backgrounds and expertise. While our approach is not sufficient to holistically assess whether the deployment of an AI system promotes equitable health outcomes, we hope that it can be leveraged and built upon towards a shared goal of LLMs that promote accessible and equitable healthcare.

Lin Ning, Luyang Liu, Jiaxing Wu et al.

Large language models (LLMs) have achieved remarkable success across various domains, but effectively incorporating complex and potentially noisy user timeline data into LLMs remains a challenge. Current approaches often involve translating user timelines into text descriptions before feeding them to LLMs, which can be inefficient and may not fully capture the nuances of user behavior. Inspired by how LLMs are effectively integrated with images through direct embeddings, we propose User-LLM, a novel framework that leverages user embeddings to directly contextualize LLMs with user history interactions. These embeddings, generated by a user encoder pretrained using self-supervised learning on diverse user interactions, capture latent user behaviors and interests as well as their evolution over time. We integrate these user embeddings with LLMs through cross-attention, enabling LLMs to dynamically adapt their responses based on the context of a user's past actions and preferences. Our approach achieves significant efficiency gains by representing user timelines directly as embeddings, leading to substantial inference speedups of up to 78.1X. Comprehensive experiments on MovieLens, Amazon Review, and Google Local Review datasets demonstrate that User-LLM outperforms text-prompt-based contextualization on tasks requiring deep user understanding, with improvements of up to 16.33%, particularly excelling on long sequences that capture subtle shifts in user behavior. Furthermore, the incorporation of Perceiver layers streamlines the integration between user encoders and LLMs, yielding additional computational savings.

Warren Morningstar, Alex Bijamov, Chris Duvarney et al.

We study the relative effects of data augmentations, pretraining algorithms, and model architectures in Self-Supervised Learning (SSL). While the recent literature in this space leaves the impression that the pretraining algorithm is of critical importance to performance, understanding its effect is complicated by the difficulty in making objective and direct comparisons between methods. We propose a new framework which unifies many seemingly disparate SSL methods into a single shared template. Using this framework, we identify aspects in which methods differ and observe that in addition to changing the pretraining algorithm, many works also use new data augmentations or more powerful model architectures. We compare several popular SSL methods using our framework and find that many algorithmic additions, such as prediction networks or new losses, have a minor impact on downstream task performance (often less than $1\%$), while enhanced augmentation techniques offer more significant performance improvements ($2-4\%$). Our findings challenge the premise that SSL is being driven primarily by algorithmic improvements, and suggest instead a bitter lesson for SSL: that augmentation diversity and data / model scale are more critical contributors to recent advances in self-supervised learning.

Yuhan Liu, Lin Ning, Neo Wu et al.

User sequence modeling is crucial for modern large-scale recommendation systems, as it enables the extraction of informative representations of users and items from their historical interactions. These user representations are widely used for a variety of downstream tasks to enhance users' online experience. A key challenge for learning these representations is the lack of labeled training data. While self-supervised learning (SSL) methods have emerged as a promising solution for learning representations from unlabeled data, many existing approaches rely on extensive negative sampling, which can be computationally expensive and may not always be feasible in real-world scenario. In this work, we propose an adaptation of Barlow Twins, a state-of-the-art SSL methods, to user sequence modeling by incorporating suitable augmentation methods. Our approach aims to mitigate the need for large negative sample batches, enabling effective representation learning with smaller batch sizes and limited labeled data. We evaluate our method on the MovieLens-1M, MovieLens-20M, and Yelp datasets, demonstrating that our method consistently outperforms the widely-used dual encoder model across three downstream tasks, achieving an 8%-20% improvement in accuracy. Our findings underscore the effectiveness of our approach in extracting valuable sequence-level information for user modeling, particularly in scenarios where labeled data is scarce and negative examples are limited.

Harrison Lee, Samrat Phatale, Hassan Mansoor et al.

Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but gathering high-quality preference labels is expensive. RL from AI Feedback (RLAIF), introduced in Bai et al., offers a promising alternative that trains the reward model (RM) on preferences generated by an off-the-shelf LLM. Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. Furthermore, we take a step towards "self-improvement" by demonstrating that RLAIF can outperform a supervised fine-tuned baseline even when the AI labeler is the same size as the policy, or even the exact same checkpoint as the initial policy. Finally, we introduce direct-RLAIF (d-RLAIF) - a technique that circumvents RM training by obtaining rewards directly from an off-the-shelf LLM during RL, which achieves superior performance to canonical RLAIF. Our results suggest that RLAIF can achieve performance on-par with using human feedback, offering a potential solution to the scalability limitations of RLHF.

Karan Singhal, Tao Tu, Juraj Gottweis et al.

Recent artificial intelligence (AI) systems have reached milestones in "grand challenges" ranging from Go to protein-folding. The capability to retrieve medical knowledge, reason over it, and answer medical questions comparably to physicians has long been viewed as one such grand challenge. Large language models (LLMs) have catalyzed significant progress in medical question answering; Med-PaLM was the first model to exceed a "passing" score in US Medical Licensing Examination (USMLE) style questions with a score of 67.2% on the MedQA dataset. However, this and other prior work suggested significant room for improvement, especially when models' answers were compared to clinicians' answers. Here we present Med-PaLM 2, which bridges these gaps by leveraging a combination of base LLM improvements (PaLM 2), medical domain finetuning, and prompting strategies including a novel ensemble refinement approach. Med-PaLM 2 scored up to 86.5% on the MedQA dataset, improving upon Med-PaLM by over 19% and setting a new state-of-the-art. We also observed performance approaching or exceeding state-of-the-art across MedMCQA, PubMedQA, and MMLU clinical topics datasets. We performed detailed human evaluations on long-form questions along multiple axes relevant to clinical applications. In pairwise comparative ranking of 1066 consumer medical questions, physicians preferred Med-PaLM 2 answers to those produced by physicians on eight of nine axes pertaining to clinical utility (p < 0.001). We also observed significant improvements compared to Med-PaLM on every evaluation axis (p < 0.001) on newly introduced datasets of 240 long-form "adversarial" questions to probe LLM limitations. While further studies are necessary to validate the efficacy of these models in real-world settings, these results highlight rapid progress towards physician-level performance in medical question answering.

Yitong Li, Dianqi Li, Sushant Prakash et al.

This work shows how to improve and interpret the commonly used dual encoder model for response suggestion in dialogue. We present an attentive dual encoder model that includes an attention mechanism on top of the extracted word-level features from two encoders, one for context and one for label respectively. To improve the interpretability in the dual encoder models, we design a novel regularization loss to minimize the mutual information between unimportant words and desired labels, in addition to the original attention method, so that important words are emphasized while unimportant words are de-emphasized. This can help not only with model interpretability, but can also further improve model accuracy. We propose an approximation method that uses a neural network to calculate the mutual information. Furthermore, by adding a residual layer between raw word embeddings and the final encoded context feature, word-level interpretability is preserved at the final prediction of the model. We compare the proposed model with existing methods for the dialogue response task on two public datasets (Persona and Ubuntu). The experiments demonstrate the effectiveness of the proposed model in terms of better Recall@1 accuracy and visualized interpretability.