Google Research

Scholarly research on generative AI often overlooks the rich cultural practices and structural conditions affecting Black communities, reducing their experiences to mere data points.

AMIE (Video) outperformed human physicians in critical clinical tasks, signaling a leap toward AI that can effectively engage in complex medical consultations.

Continuous state optimization can slash movement costs by over 300% while maintaining performance in complex learning systems.

MameLoshnLM not only sets a new standard for Yiddish NLP but also reveals the inadequacies of multilingual models in handling linguistically rich yet underrepresented languages.

AMS reveals that safety training modifications can significantly alter the activation landscape of language models, impacting their compliance with safety protocols.

Foundation model agents can achieve stable cooperation in social dilemmas by inferring behavioral similarities, defying classical game theory's expectations of mutual defection.

Long-term interactions with language models could lead to significant cognitive and emotional changes in users that short-term evaluations miss entirely.

No randomized polynomial-time algorithm can overcome the condition-number barrier in sparse least-squares optimization, confirming a long-standing conjecture.

Safety fine-tuning in LLMs not only suppresses self-attribution of consciousness but also inadvertently erases culturally significant beliefs about non-human entities and spirituality.

HA-MoE achieves significant gains in user engagement by effectively managing the complexities of heterogeneous content in recommendation systems.

Aligning luminance histograms can transform diffusion models into powerful HDR generators without any retraining, achieving remarkable fidelity and detail.

Achieving nearly linear sample complexity for distribution learning is possible by leveraging hierarchical comparability in query families.

Uncertainty-driven exploration can significantly boost new release visibility in music recommendations, but at the cost of user engagement and diversity.

PoTRE's innovative use of diverse reasoning agents leads to a remarkable 49.92% accuracy on Humanity's Last Exam, setting a new benchmark for LLM performance in complex reasoning tasks.

DeepDebug achieves a 32% improvement in task recovery accuracy, showcasing a powerful new approach to debugging LLM agent failures.

μSTM achieves high performance without sacrificing safety or usability, allowing for general data types and deferred aborts in transactions.

Mure achieves a 44.54% reduction in mutation testing costs without sacrificing accuracy, challenging the trade-off between efficiency and mutation score fidelity.

Ignoring the social dynamics that led to past disasters could doom AI systems to repeat history's mistakes.

HandFlow reduces world-space pose error by over 30% while reconstructing 150 frames at 47 fps, setting a new standard for 4D hand recovery.

Temperature sampling can enhance mode diversity in diffusion models without sacrificing sample quality, thanks to a novel variance-corrective technique.

Simple ensemble methods leveraging rich textual context can outperform state-of-the-art multimodal forecasting approaches on a new benchmark, TimesX, revealing hidden vulnerabilities in existing evaluations.

A pathwise approach to change detection reveals that continuous transport in feature space significantly enhances the model's ability to capture and interpret temporal changes.

The query complexity of active learning algorithms can be dramatically influenced by the graph's vertex expansion, revealing a new dimension in adversarial robustness.

CollabEval slashes evaluation uncertainty, achieving more accurate model assessments with less data by exploiting historical performance correlations.

Gemma 4's unified architecture and reasoning mode enable it to outperform larger models in human-rated tasks while maintaining high efficiency.

Bi-NAS boosts recommendation accuracy while transforming user explanations into clear, personalized insights that enhance trust and engagement.

Task-specific audio perturbations can boost model accuracy by over 6% and reduce hallucinations in large audio-language models.

AugSplat boosts reconstruction quality in sparse-view 3D vision by leveraging synthetic views from neural radiance fields, achieving real-time performance without sacrificing accuracy.

RLMF not only boosts LLMs' ability to accurately express uncertainty but also enhances their self-assessment capabilities, fundamentally reshaping their trustworthiness.

LLMs falter on Romanized Code Mixing tasks, revealing a critical gap in their multilingual instruction-following abilities.

LiDAR-based 3D object detectors can be compromised by targeting just a few critical spatial regions, revealing a significant structural vulnerability.

A new AI tool can catch 34% more mathematical errors in scientific papers, transforming the peer review landscape.

Contrastive learning can significantly enhance speech quality assessment models without the computational burden of multi-stage training.

Achieving accurate pose estimation with rolling shutter cameras using just seven correspondences could revolutionize real-time applications in consumer devices.

Go's unique structural subtyping and generics can now be formally captured without sacrificing runtime efficiency or compatibility with existing compilation practices.

Language models encode knowledge in a task-specific manner, leading to inconsistent retrieval of facts across different tasks.

COrigami shows that AI can effectively merge strict geometric requirements with subjective artistic aesthetics, transforming the origami design process.

Jointly verifying user intent and response harm can reduce attack success rates to just 4.1%, setting a new standard for LLM safety.

Triangle splats from video diffusion latents yield superior geometric accuracy and visual quality, challenging the dominance of volumetric 3D Gaussians in scene generation.

Directional sharpness outperforms traditional metrics in predicting model generalization, making it a game-changer for model certification.

Cross-lingual exploration can unlock hidden knowledge in LLMs, improving factual recall and consistency across 17 languages.

Multi-vector embeddings can express complex similarities that single-vector embeddings simply cannot, even with optimal representation sizes.

Achieving optimal regret bounds without the complexity of count-based uncertainty estimates could revolutionize exploration strategies in reinforcement learning.

WEQA achieves a 24% accuracy boost in wearable health question answering by dynamically adapting to the complexities of sensor data and user queries.

Ignoring the interplay between engineering, business, and legal perspectives can doom privacy-enhancing technologies to failure in software systems.

RubricsTree transforms the evaluation landscape for personal health agents, achieving expert alignment and significant performance gains while addressing the scalability challenge in clinical deployment.

VisualClaw slashes API costs by 98% while boosting accuracy, transforming how VLMs can operate in real-time environments.

A novel auditing framework reveals that synthetic data can leak private information without model access, challenging assumptions about data privacy in generative AI.

LLM agents struggle to uncover hidden environments, with performance plummeting as task complexity increases, revealing fundamental limitations in their interactive reasoning capabilities.

Delayed feedback in linear bandits can fundamentally alter regret dynamics, revealing that loss-dependent delays are significantly more challenging than in multi-armed bandits.

No single model dominates video embedding tasks, revealing stark contrasts in performance based on modality and task type.

Real-time LLM-generated user personas can dramatically enhance viewer engagement by dynamically balancing existing interests with new content recommendations.

PI-Hunter uncovers hidden prompt injection vulnerabilities in LLM agents that traditional defenses miss, revealing a critical gap in current security practices.

APEX reveals that optimizing data alongside prompts can boost LLM performance by over 11% while significantly reducing wasted compute resources.

LLMs reveal surprising strengths and weaknesses in analyzing security logs, with performance heavily influenced by model design choices.

Redefining data work through a reparative lens reveals the urgent need to prioritize the voices of those harmed by online systems, challenging existing norms of accountability in AI.

In high-stakes health contexts, stakeholders demand that trustworthiness in AI systems be inspectable, not just asserted, reshaping how we design health information tools.

A strategic messaging shift on Google Search reduced CSAM-related queries by 3.8%, effectively redirecting some users towards therapeutic resources.

MResOpt achieves significantly lower high-priority constraint violations in constrained optimization tasks while remaining computationally efficient, revolutionizing how we approach complex optimization problems.

RLNS turns a classic heuristic into a powerful MCMC sampler, enabling efficient combinatorial optimization without the need for exact solutions.

Non-private synthetic data can effectively transfer knowledge from original corpora, while state-of-the-art DP methods often fail to do so, even at high privacy levels.

VLMs struggle with procedural 3D modeling, often producing flawed outputs due to API mismatches and geometric disconnections, but performance can be significantly boosted through iterative refinement.

Ditch the brittle code synthesis and noisy gradients: LiveSVG unlocks high-quality SVG animations by directly fitting vector graphics to reference videos generated from motion prompts.

LLM-powered honeypots can trick even frontier models into longer interactions than rule-based systems, all while costing less to run.

PARCEL redefines visual tokenization, achieving superior efficiency and performance by dynamically anchoring feature extraction to spatial pool tokens.

Splitting attention and feedforward networks onto separate GPUs can unlock 4x higher MoE LLM throughput, but only if you carefully tune the GPU partitioning strategy based on the workload.

Why pick just one token mixer when you can have them all, dynamically switching between attention and linear recurrences for optimal efficiency and performance?

LLMs alone can't reliably retrieve actionable data from the web, with agents relying on semantic metadata achieving 65% higher precision in finding FAIR-compliant datasets.

Humans miss 3.9% of opportunities to leverage correct AI suggestions while also over-relying on misleading outputs, highlighting critical gaps in trust and decision-making in human-AI collaboration.

Achieving provable, non-asymptotic guarantees for optimizing complex multi-label metrics like F-measure is now possible with a new family of algorithms that decompose exactly for $O(l)$ time complexity.

Imagine telepresence where your avatar convincingly blends into any environment, relit in real-time based on the scene's actual lighting, all from a single headset.

Gemini Embedding 2's unified multimodal embeddings beat specialized models across diverse tasks and even generalize zero-shot to niche fields like astronomy and culinary arts.

Even the best LLM judges miss cultural faux pas that are obvious to locals, achieving only 52% F1 score on a new benchmark.

Bandit feedback doesn't have to cripple learning: a new "bandit DS dimension" reveals how to achieve near-optimal sample complexity in multiclass PAC learning, even when you only know if you're right or wrong.

AI-driven scientific discovery is closer than you think, but current systems still struggle with reproducibility, cross-domain robustness, and accountable scientific closure.

You can slash the compute cost of visual geometry transformers by 85% without sacrificing accuracy by intelligently pruning redundant tokens across frames and within layers.

Graph transformers can be fundamentally limited by their tokenization strategy, as some tokenizations provably preclude efficient learning of structural representations realizable with other tokenizations.

Training a foundation model on a trillion minutes of wearable sensor data unlocks surprisingly accurate predictions across a wide range of health conditions, even with limited labeled data.

AI can now autonomously solve open math problems, cracking 9 Erdős problems and 44 OEIS conjectures at a reasonable cost.

Instead of creating new AI companions from scratch, Deco shows how to breathe new life into cherished physical objects by giving them a digital voice and personality powered by LLMs.

LLMs' persistent hallucinations aren't just about lacking knowledge, but about lacking the self-awareness to know what they *don't* know, suggesting uncertainty expression is key to building trustworthy AI.

Forget handcrafted prompts: a hierarchical multi-agent framework turns diffusion models into coherent storytelling engines by globally optimizing for semantic coherence.

Current remote sensing change captioning datasets miss fine-grained localized semantic reasoning, but RSRCC fills this gap with 126k change-specific questions.

Stop penalizing your ANN search algorithms for failing to retrieve irrelevant neighbors – Semantic Recall offers a more nuanced and effective way to measure retrieval quality.

LVLMs can self-detect and correct object hallucinations by focusing on specific image regions, offering a simple, training-free fix.

GAAP offers a deterministic, trust-minimized approach to AI agent security, safeguarding user data even when models are compromised or prompts are injected.

Multilingual LLMs exhibit a surprising "American bias," even when prompted in other languages, and instruction tuning makes it worse.

ZKP proving, previously bottlenecked by MSM and NTT operations, can now achieve up to 10x higher throughput on TPUs thanks to a novel framework that reformulates ZKP kernels for AI-ASIC execution.

FUSE achieves verification quality on par with semi-supervised methods, all without needing any labeled data.

Debloating tools, intended to shrink code and improve security, can actually *add* code or remove essential functionality, with dynamic methods being overly aggressive and static methods overly conservative.

RosettaSearch recovers up to 68% more structural fidelity in protein designs, transforming how we optimize sequences beyond traditional single-pass methods.

Generating consistent visual narratives is now possible: CANVAS outperforms existing methods by explicitly planning character, background, and scene continuity across multiple shots.

Reconstructing dynamic hand-object interactions from monocular video can be 6x faster and significantly more accurate by ditching heavy neural representations for a revived Sum-of-Gaussians approach.

Ethics interventions in AI development often fail because practitioners don't trust them – here's a breakdown of why, and how to fix it.

Google developers are spending less time debugging integration tests thanks to an LLM that diagnoses failures with 90% accuracy.

Unpacking Google's AI literacy partnerships reveals the surprising complexities of aligning research, industry, and public needs.

Ditch imperative robot programming and embrace the elegance of logic: control swarms with declarative code.

Forget KL divergence – this work shows you *can* reliably evaluate generative models with finite samples, but only if you use the right metric (IPMs with bounded test classes).

LLMs can now generate more relevant and factual movie recommendations by dynamically bridging retrieval and generation with a novel reinforcement learning approach.

CGRA performance jumps by 2.7x thanks to NEURA, a compilation framework that elegantly transforms control flow into dataflow.