If you're struggling with the credibility and grounding of your LLM's generated text, our new approach, CaLM: Contrasting Large and Small Language Models to Verify Grounded Generation (ACL '24), might be the solution you've been looking for. 💡 Paper: arxiv.org/abs/2406.05365 1/n

🚀 New paper alert! Want to make your LLM's generation more credible and grounded on pieces of evidence? Our CaLM algorithm, presented at k'24, leverages smaller LMs to achieve just that **unsupervisedly**! (1/n)

Why LLMs lost in the middle❓ 💡LLMs exhibit U-shape positional attention bias that dominates their generation behavior (often using leading/ending contexts in the response) 🚀By modeling and removing such bias, we hugely improve LLMs RAG performances! 📜: arxiv.org/abs/2406.16008

Swarm intelligence is the key! Check out our work led by our awesome intern Shangbin Feng at Google AI Cloud AI Research.

🚨 Excited to share our latest work Google DeepMind: OmnixR, an evaluation suite for evaluating the reasoning of Omni-Modality Language Models' (OLMs) across modality. We observe the significant reasoning performance drops of all the SoTA OLMs on other modalities compared to the

Searching for the ultimate LLM Knowledge distillation! Want one that excels in both task-specific and task-agnostic settings. Could it outperform others on varying data sizes and model initializations? Our Speculative Knowledge distillation might be the answer🚀 Google

New from University of Washington and Google DeepMind: Model Swarms, a collaborative search algorithm that adapts LLM experts to single task, multi-task domains, and reward models via swarm intelligence. Talk to the team Zifeng Wang Chen-Yu Lee Yejin Choi tsvetshop here! alphaxiv.org/abs/2410.11163…

Our team (Google Cloud AI Research: research.google/teams/cloud-ai/) is seeking PhD student researchers/interns to work on LLM-related topics (agent, reasoning, RAG, data synthesis, etc.). If interested, please fill in this form: forms.gle/Cia2WGY94zTkpP…. Thank you and plz help RT!

🚨 Reverse Thinking Makes LLMs Stronger Reasoners We can often reason from a problem to a solution and also in reverse to enhance our overall reasoning. RevThink shows that LLMs can also benefit from reverse thinking 👉 13.53% gains + sample efficiency + strong generalization!

Happy to share that RevThink has been accepted to #NAACL2025 main conference! 🎉We also release the code and data 👇🧵 RevThink shows that LLMs can also benefit from reverse thinking (like we often do) 👉13.53% gains on 12 datasets (including MATH, ARC, ANLI, etc) + sample

Shangbin is super brilliant:) Great pleasure working with you at Google!

👀 How to effectively leverage the expertise of diverse models? ✨ Optimize graphs of LLMs with swarm intelligence! 👉🏻 Introducing Heterogeneous Swarms, jointly optimizing the roles and weights of multi-LLM systems for collaborative gains! 📄 Paper: arxiv.org/abs/2502.04510

I will be presenting ✨Reverse Thinking Makes LLMs Stronger Reasoners✨at #NAACL2025 ! We show that LLM can also benefit from reverse thinking -- a technique we often use to reason from a problem to a solution: - Improvements across 12 datasets - Outperforms SFT with 10x more

Now accepted at ICML 2025

AlphaEvolve, our new Gemini-powered coding agent, can help engineers + researchers discover new algorithms and optimizations for open math + computer science problems.  We’ve used it to improve the efficiency of our data centers (recovering 0.7% of our fleet-wide compute

🧠🚀 Excited to introduce Supervised Reinforcement Learning—a framework that leverages expert trajectories to teach small LMs how to reason through hard problems without losing their minds. 🤯 Better than SFT && RLVR. Read more: huggingface.co/papers/2510.25… #llms #RL #reasoning

Your agents are smarter and more efficient with Budget Tracker!

Budget-Aware Tool-Use Enables Effective Agent Scaling

Google introduces Budget Tracker for smarter AI agents Current LLM agents waste tool-call budgets. This work unveils Budget Tracker and BATS, enabling agents to dynamically adapt planning based on remaining resources.

At #NeurIPS2025 in San Diego from Dec 1 to 7 ✈️ Happy to meet old friends and make new ones! Looking forward to exchanging ideas on effective and efficient LLM agents, multi-LLM/agent systems, post-training and continual learning for LLMs, and much more! Also, our team, Cloud