FAQ

Aya began with the bold vision of creating a language model that could serve a broader global audience, and this vision became a reality with the launch of Aya 101 in February, 2024. Aya 101 was a groundbreaking open-source, massively multilingual generative large language model that supported an unprecedented 101 languages. This project was made possible by the collaboration of 3,000+ researchers from 119 countries, making it one of the largest open science efforts in ML to date. Together, we created a model that not only outperformed existing open-source models but also expanded coverage to over 50 previously underserved languages. We open-sourced the Aya model, dataset, and rare human-curated annotations, ensuring that AI technology can be more inclusive and effective for diverse language and cultural communities. Along with being a technical accomplishment, in many ways Aya was a protest, challenging who gets to participate in scientific breakthroughs. Explore the storys of Aya and here from its creators in our short documentary, The Journey of Aya.

Building on the success of Aya 101, we continued our journey with the release of Aya 23, taking our multilingual language modeling capabilities to the next level. This advanced 8B and 35B open weights model experimented with the depth, allocating more capacity to a smaller set of languages during pre-training. Aya 23 set new benchmarks in multilingual generative AI, achieving state-of-the-art performance and serving nearly half of the world's population with its enhanced language modeling capabilities.

The Aya family of models, including Aya 101 and Aya 23, showcases our unwavering commitment to treating more languages as first-class citizens. We believe that breakthrough research should be accessible to the wider research community, and that's why we've made our models available for experimentation, exploration, and safety auditing.

Cohere For AI is a non-profit research lab that seeks to solve complex machine learning problems. We support fundamental research that explores the unknown, and are focused on creating more points of entry into machine learning research. Learn more at our website.

We're excited to support research projects that advance multilingual, and make use of Aya 23 (8B or 35B), Aya 101, or other multilingual tools. Check out this presentation for some examples, and be in touch on Discord in #project-idea-sharing to get feedback on ideas. 

There are many ways to connect with prospective teammates:

Sign-up Sheet - share your research ideas in the sign-up sheet, and browse ideas from others. Look up their Discord handles to be in touch and express your interest in joining their crew.

Discord - Do you have a project idea and want to share it or get some feedback on it? Share more details in #project-idea-sharing. Looking for a team? Introduce yourself in #project-idea-sharing, including what kind of project you may want to work on, and scroll through the intros of others to find collaborators.

Crew Connections meetings - join either of our crew connections meetings. Here we will introduce the Expedition more thoroughly and there will be plenty of time to connect with others who are looking to form a team. Details on the Home page.

We are committed to supporting all participants throughout the expedition. This includes:

• Facilitating connections: We'll help you find potential collaborators to expand your network and expertise to enhance your project.

• Cohere API credits: To support your project development, we are offering API credits. This will provide access to Aya 23 models to take your project to the next level.

• Technical Guidance: Our team will provide the initial technical guidance to help you get started and give timely feedback to help you make progress during the expedition.

• Regular check-ins: We want to ensure your progress and address any concerns, so we'll schedule regular check-ins. This will help keep everyone on track and allow for any necessary adjustments.

• Closing event & prizes: Finally to celebrate your hard work, we will host a closing event with exciting prizes.

We want to ensure that you have the resources and support necessary to make the most of this expeditionary journey.

We aim to provide comprehensive support, but there are a few areas that will not be covered during the competition:

• Compute support for training runs: Participants will need to utilize their own computational resources or cloud services for any machine learning model training and experimentation.

• Code debugging: While we offer technical guidance and tools, we will not be able to provide individual code debugging support. Participants are expected to troubleshoot and resolve coding issues independently or through their collaborators.

• Project idea generation: The competition encourages creativity and initiative. Thus, we will not provide specific project ideas or suggestions. Participants should develop their own unique project proposals and see them through to completion.

• Financial support beyond the stated prizes: Aside from the Cohere API credits and prizes offered, there will be no additional financial support provided for project development or related expenses.

• Post-expedition support: The expedition concludes with the closing event and prize distribution. We will not be able to provide extended support or resources beyond the competition timeframe.

The expedition will run till end with the closing ceremony on August 28.

The focus of this expedition is on kickstarting innovative research initiatives and fostering collaboration, rather than building finished products. We encourage participants to think big and explore new ideas, even if they may not result in a finished deliverables within the 6-week timeframe. 

Our expectations are that teams will work collaboratively to develop unique and impactful research projects focussed on accelerating multilingual AI. The specific deliverables will vary depending on the nature of each team's project but could include technical blog showcasing findings or proof-of-concept demonstrations. Each team will be expected to report on weekly progress and showcase their work as part of our closing ceremony. 

We hope that the connections and collaborations formed during the competition will extend beyond the 6 weeks and contribute to long-term research partnerships.

Anyone is welcome to participate in the expedition. You do not need to have any kind of prior research experience to participate. You also don't need to be a former Aya collaborator to participate in the Aya expedition. 

Check out our work on the Aya project to get inspiration for project ideas. 

Our team is in the process of designing exclusive Expedition Aya swag for participants who complete the expedition and those who produce top projects, evaluated on metrics such as creativity, potential impact, and scientific integrity. Stay tuned for more information!