Connect 2025

North Mini Code. Cohere's first model for developers.

Learn more

Dec 08, 2025

3 minutes read

Connect 2025: Why Collaboration is the Ultimate Research Superpower

!Blog Post Featured Image

If Connect 2025 proved anything, it’s that collaboration is not just a nice-to-have, but a research superpower. In a world where AI problems are increasingly complex and high-stakes, no single researcher or team can tackle them alone. Over three energetic days, Cohere Labs’ open science community met, shared, questioned, and together showed how curiosity, openness, and collective effort can transform ambitious ideas into breakthroughs.

This year’s Connect conference was a celebration of rigorous research, unexpected partnerships, and the creative collisions that happen when curious people gather with a shared purpose…to advance machine learning through open science.

###### Collaboration matters for every researcher

Collaboration benefits everyone, from those just starting their research journey to seasoned experts shaping the field.

• Early-stage researchers gain access to mentorship, diverse perspectives, and opportunities to contribute to projects that would be impossible alone.
• Late-stage researchers can scale ambitious projects, test bold hypotheses, and challenge assumptions with fresh eyes.

At Connect 2025, this principle came alive. Joelle Pineau, Cohere’s Chief AI Officer, and Marzieh Fadaee, Head of Cohere Labs, reminded us that bold research thrives when more diverse voices are invited to the table. Their message set the tone - open collaboration multiplies impact.

###### Sparking innovation

Innovation rarely happens in isolation. Lightning talks at Connect 2025 showcased how collaboration fuels creativity. Key examples included:

• Aya: What began as an exploratory effort, quickly evolved into the largest open science collaboration in machine learning, bringing together a community of researchers from around the world to create a powerful foundation for future innovation in a global technology.
• Global MMLU :We started this cross-institutional collaboration with Hugging Face, EPFL, NUS, AI Singapore, Mila, and other institutions. To identify culturally sensitive samples and improve translation quality, the project later expanded to our open science community, engaging contributors across 42 languages.
• Maya: Born from Expedition Aya (a six-week open build sprint where researchers from around the world came together to experiment boldly), Maya emerged as a collaborative, open-source multilingual VLM project dedicated to advancing culturally inclusive and toxicity-aware vision-language modeling.

Following the talks, we hosted AMA sessions with the authors where honest questions met practical experience, often sparking surprising insights and new collaborations.

###### Collaboration scales ambition

Some goals are simply too big for one person or lab to tackle alone. In this fireside chat, Cohere co-founder Ivan Zhang, reflected on how Cohere Labs began with the belief that collaboration accelerates discovery and how that vision continues to guide efforts like building inclusive global benchmarks and multilingual datasets.

During our second lineup of lightning talks, we saw collaboration expressed in countless forms:

• Afri-Aya: Born from Expedition Aya this project began as a community effort to curate a multilingual image dataset that covered 13 major African languages with AI-powered categorization.
• Do Biased Models Have Biased Thoughts: The team behind this paper met through our open science community at one of our Research Connection sessions and investigated whether LLM internal reasoning reflects the biases in their outputs.

Large-scale collaborations accelerate discovery, enabling progress on high-impact problems that would otherwise be impossible for isolated researchers. Collaboration doesn’t just make research easier, it makes the impossible achievable.

###### From ideas to action

Our final day featured a keynote from Shayne Longpre, PhD, AI Research Scientist at MIT, who presented a detailed roadmap for successful research collaborations.

Key lessons on successful collaboration in AI research included:

• Start with a clear idea, circulate a one-pager to collaborators and critics.
• Recruit broadly across geographies, institutions, and experience levels.Set clear incentives around authorship, credit, and ownership.
• Communicate early and often about timelines, roles, and expectations.
• Document decisions, links, and contributions as the project grows.

We closed with final sessions hosted by Cohere Labs researchers, complete with interactive boards and more, “Wait, what if we tried…” moments than we could count. These sessions captured the true spirit of Connect 2025: curiosity in action, ideas sparking off one another, and collaboration turning imagination into possibility.

Beyond the talks and presentations, the conference demonstrated that research breakthroughs don’t happen in isolation, they happen when diverse minds come together, ask bold questions, and build on each other’s insights. Connect 2025 was just the beginning, and the next wave of innovation will come from the projects, conversations, and collaborations seeded here.

###### Ready to start your own collaboration? Check out these resources:

Learn more about our Open Science Community and join us on Discord. This is the perfect place to help you start your research collaboration.

Learn more about the Cohere Labs Catalyst Grants which provide researchers and developers with free access to the Cohere API to support their projects...