Google Research

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

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

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

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

Training LLMs with a compact rolling memory can lead to more robust reasoning, outperforming models that rely on full historical context.

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.

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

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.

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.

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

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.

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.

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.

Semantic search across hundreds of millions of clinical notes is not just feasible, but can slash chart review times by up to 89% while maintaining accuracy.

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

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.

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).

Safety fine-tuning might inadvertently be stripping LLMs of their ability to understand non-human minds and entertain spiritual beliefs, even while preserving Theory of Mind.

Despite the effort required, Android developers overwhelmingly support platform-level changes to combat fingerprinting, suggesting a path to enhanced user privacy through collaborative platform-developer initiatives.

LLM-powered diagnostic AI is ready for prime time: a real-world clinical trial shows it's safe, patients love it, and doctors find it useful.

Forget catastrophic forgetting: this function-preserving expansion method lets you fine-tune without sacrificing pre-trained knowledge, matching full fine-tuning performance at a fraction of the cost.

Finally, a framework to quantify AI's cultural intelligence, moving beyond ad-hoc cultural benchmarks to a systematic, extensible, and theoretically grounded approach.

Recurrent models can now achieve Transformer-competitive performance on recall-intensive tasks, thanks to a simple memory caching mechanism that grows memory capacity with sequence length.

Randomly masking parameter updates in RMSProp delivers state-of-the-art LLM training performance, revealing a surprisingly effective form of geometric regularization.