Research
A paper is listed when one of its authors is affiliated with the lab. Data: OpenAlex.
Self-Improving Embodied Foundation ModelsAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios Using ConcordiaAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025Too Late to Recall: Explaining the Two-Hop Problem in Multimodal Knowledge RetrievalAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMsGoogle DeepMind · neural information processing systemsJan 1, 2025Bigram Subnetworks: Mapping to Next Tokens in Transformer Language ModelsGoogle DeepMind · neural information processing systemsJan 1, 2025The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT ImprovementsGoogle DeepMind, Meta · neural information processing systemsJan 1, 2025Isotropic Noise in Stochastic and Quantum Convex OptimizationAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025Generalizable Reasoning through Compositional Energy MinimizationGoogle DeepMind · neural information processing systemsJan 1, 2025Training Language Models to Generate Quality Code with Program Analysis FeedbackAnthropic, Google DeepMind, Microsoft · neural information processing systemsJan 1, 2025Exploration from a Primal-Dual Lens: Value-Incentivized Actor-Critic Methods for Sample-Efficient Online RLAlphabet, Google DeepMind, Meta · neural information processing systemsJan 1, 2025Imagined AutocurriculaGoogle DeepMind · neural information processing systemsJan 1, 2025Robust and Diverse Multi-Agent Learning via Rational Policy GradientGoogle DeepMind · neural information processing systemsJan 1, 2025Kernel Density Steering: Inference-Time Scaling via Mode Seeking for Image RestorationAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025A Clean Slate for Offline Reinforcement LearningGoogle DeepMind · neural information processing systemsJan 1, 2025On Relation-Specific Neurons in Large Language ModelsGoogle DeepMind · Conference on Empirical Methods in Natural Language Processing (EMNLP)Jan 1, 2025How Well Can Reasoning Models Identify and Recover from Unhelpful Thoughts?Google DeepMind · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)Jan 1, 2025OVFact: Measuring and Improving Open-Vocabulary Factuality for Long Caption ModelsGoogle DeepMind · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)Jan 1, 2025Internal states before wait modulate reasoning patternsAlphabet, Google DeepMind · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)Jan 1, 2025Infinite Neural Operators: Gaussian processes on functionsGoogle DeepMind · neural information processing systemsJan 1, 2025AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforced Self-PlayGoogle DeepMind · neural information processing systemsJan 1, 2025Dense SAE Latents Are Features, Not BugsGoogle DeepMind · neural information processing systemsJan 1, 2025Tracing the Representation Geometry of Language Models from Pretraining to Post-trainingAlphabet, Cohere, Google DeepMind · neural information processing systemsJan 1, 2025Reverse Engineering Human Preferences with Reinforcement LearningCohere, Google DeepMind · neural information processing systemsJan 1, 2025Hierarchical Retrieval: The Geometry and a Pretrain-Finetune RecipeAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025Plasticity as the Mirror of EmpowermentAlphabet, Google DeepMind · neural information processing systemsJan 1, 2025