Cohere Labs

Learning is the goal. 📚 Community lead, Sree Harsha Nelaturu reminds us that impactful research isn’t about chasing the most complex ideas— it’s about following what excites you and growing through the process. Join our Open Science Community: https://lnkd.in/gTbtNjqf

“Individually, we are one drop. Together, we are an ocean.” - Ryunosuke Satoro ✨ Cohere Labs is excited to announce Connect - a 3-day virtual conference celebrating the power of collaboration in open science! Join us for inspiring keynotes, lightning talks, and interactive sessions that bring together curious minds from around the world. Throughout the conference, we’ll: 🔬 Showcase cutting-edge research 💡 Highlight meaningful collaborations 🤝 Inspire new partnerships We're excited to hear from speakers including Ivan Zhang, Joelle Pineau, Marzieh Fadaee, Shayne Longpre and 20+ other presenters who will share insights on open science, collaborative research, and community-driven innovation. Learn more and register now: https://lnkd.in/gcS6E9DW

Our AI Safety and Alignment group is excited to host Hashim Ali for a presentation on "Audio Antispoofing in the Age of Hyper-Realistic Deepfakes: Challenges, Datasets, and Robustness" on October 30th. Thanks to our community leads Abrar Rahman and Alif Munim for organizing this event! 👏 Learn more: https://lnkd.in/gMTMd2kE

The Art of Asking: Multilingual Prompt Optimization for Synthetic Data 🌍. In multilingual synthetic data generation, relying solely on translation often produces prompts that feel unnatural, Western-centric, or linguistically flat. Most existing methods focus on optimizing completions—the model’s responses—while overlooking the quality of the prompts themselves. But prompts shape the data we generate: Completions are only as diverse and representative as the inputs they are conditioned on. Our work introduces a prompt-focused paradigm, transforming translated prompts along three key dimensions: naturalness, cultural authenticity, and complexity. By treating prompts as adaptable, not static, we create richer, more realistic inputs that better mirror human data. These simple transformations yield consistent improvements across languages and benchmarks—especially for open-ended tasks—enhancing fluency, diversity, and challenge in generated outputs. Sometimes, the way we ask a question really does make all the difference 💡✨ Led by: David Mora, Viraat Aryabumi, Wei-Yin Ko, Sara Hooker, Julia Kreutzer, and Marzieh Fadaee. 📜Paper link: https://lnkd.in/gAHiWmuU

I’m very happy to share our latest work: The Art of Asking: Multilingual Prompt Optimization for Synthetic Data 🎨🌍. In multilingual synthetic data generation, relying solely on translation often produces prompts that feel unnatural, Western-centric, or linguistically flat. Most existing methods focus on optimizing completions—the model’s responses—while overlooking the quality of the prompts themselves. But prompts shape the data we generate: Completions are only as diverse and representative as the inputs they are conditioned on. Our work introduces a prompt-focused paradigm, transforming translated prompts along three key dimensions: naturalness, cultural authenticity, and complexity. By treating prompts as adaptable, not static, we create richer, more realistic inputs that better mirror human data. These simple transformations yield consistent improvements across languages and benchmarks—especially for open-ended tasks—enhancing fluency, diversity, and challenge in generated outputs. Sometimes, the art of asking the right question really does make all the difference 💡✨ To check more about our work check out our paper: (https://lnkd.in/eJvdWT3X) and special thanks to my mentors and collaborators Marzieh Fadaee, Julia Kreutzer, Viraat Aryabumi, Sara Hooker, Wei-Yin Ko ✨✨✨

Don't forget to join us tomorrow, October 23rd for a session with Calarina Muslimani on "Towards Improving Reward Design in RL: A Reward Alignment Metric for RL Practitioners" Learn more: https://lnkd.in/gHSmP_u2

Share Your Research with Our Open Science Community! 🎙️ Our open science community thrives on collaboration and knowledge-sharing in the various fields of machine learning. We’d love to hear from you if: 🔬You've conducted some exciting research and want to share your findings 📚 You're currently a student and looking for the opportunity to present your latest discoveries 👩🏫 You’re an ML industry expert ready to share your knowledge and experience Submit your interest today: https://lnkd.in/gkXXsNWh

Be sure to tune in tomorrow, October 21st as we welcome Zhiwen(Aaron) F. for a session on "VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction." Learn more: https://lnkd.in/g_FP-4FV

Researchers and the public see AI in health care differently. 🏥🤖 While researchers focus on its potential to improve lives, many people remain uncertain or uneasy. Dr. Satish E. Viswanath explores how we can start bridging that divide: https://lnkd.in/gG-cXXyB