We've published a new policy primer: The Limits of Thresholds. Here's a summary: Some AI governance frameworks specify risk thresholds based on training compute, but growing evidence suggests that increased training compute does not necessarily equate to increased risk. 📈❔ Research shows that many other factors beyond compute influence the risks of an AI model, such as the characteristics of training and inference datasets, a model’s deployment context, performance and safety optimization techniques. Plus, increasingly smaller models match or outperform larger ones. The limitations of compute thresholds have several consequences for AI governance, in that they could: ⚖️ Underestimate risks of smaller models & overestimate risks of larger ones. 🎲 Incentivise threshold gaming. 🔀 Divert resources away from effective risk governance. To address these limitations, policymakers could consider: 🤝 Adopting alternative or complementary approaches to compute for setting risk thresholds. 📊 Developing dynamic rather than static thresholds. 🧮 More clearly defining approaches to calculating compute. Read the full primer here: https://lnkd.in/eBAiZfSQ
Cohere For AI
Non-profit Organizations
We’re a non-profit ML research lab and community exploring the unknown, together.
About us
Who we are 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. Our community is a space where researchers, engineers, linguists, social scientists and lifelong learners connect and collaborate with each other. We come together from all over the world and welcome you whether you are a mentor, dropout, just getting started, PhD, masters, undergraduate, unaffiliated, industry, academic or not really sure. We are excited to support community-driven research and to be shaped by our members' interests. Where we’ve come from In 2017, a team of friends, classmates, and engineers started a distributed research collaboration, with a focus on creating a medium for early-career AI enthusiasts to engage with experienced researchers – they called it “for.ai.” Two of those co-founding members, Aidan Gomez and Ivan Zhang, later went on to co-found Cohere, and many of the founding members went on to do exciting things (pursuing PhDs, working at industry and academic labs). At the time, For AI was one of the first community-driven research groups to support independent researchers around the world. Today, Cohere is proud to reintroduce For AI as Cohere For AI, a dedicated research lab and community for exploring the unknown, together.
- Website
-
https://cohere.com/research
External link for Cohere For AI
- Industry
- Non-profit Organizations
- Company size
- 2-10 employees
Updates
-
August 13th, join Mia (Miao) Zhang and our community-led Computer Vision group for a presentation on "Common-Sense Bias Discovery and Mitigation for Classification Tasks." 😎 Learn more: https://lnkd.in/gNSVxkmd Thank you Ameed Taylor and Benedict Emoe-kabu for putting this talk together! 💪
-
-
Tomorrow, our community-led ML for Good group will have Andrew Trask presenting. Be sure to check it out! ✨ Learn more: https://lnkd.in/gMtff4K2
August 6th, you can catch Andrew Trask with our ML for Good group who will be presenting "The future of AI Infrastructure: training models while preserving data privacy." 🤳 Learn more: https://lnkd.in/gMtff4K2 Thank you Jen Iofinova and Kate Gamper for organizing this event! 👏
-
-
The Association for Computational Linguistics 2024 conference in Bangkok, Thailand is less than a week away and we are excited to share what our Cohere and C4AI teams will be presenting. 🙌 Check out where we are throughout the conference here:
-
-
Cohere For AI reposted this
دوستان اگه می تونین لطفا یکم زمان به این برنامه اختصاص بدین.🙏 در دسترس بودن این بنچ مارک ها در همه جا به کارمون خواهد اومد، از انتخاب راحت تر مدل برای کار های پایین دست خودمون، تا تشویق شرکت هایی مثل خود Cohere برای تولید LLM هایی که توی فارسی هم مهارت خوبی دارن، چون واقعا کار اون ها رو هم راحت تر می کنه. به شخصه سختی هایی که از عدم وجود evaluation مناسب به وجود میاد رو دست اول چشیدم.
3 days left to validate translations in your language an collaborate toward making multilingual benchmarks more equitable. Thanks to the 154 people who have contributed, and particularly to the 40 who have done over 100 validations! 🏁 8/39 languages complete. 🎊 Vietnamese, Portuguese, & Amharic are all over 1/2 way done! 🏃♂️ German, Indonesian, Sinhala, Turkish, Nepali, Swedish, and others have progress, but will need more help to cross the finish line before our Friday deadline! Join us: https://lnkd.in/gPuK57uW Track the progress: https://lnkd.in/gCkzNBxb
-
-
Cohere For AI reposted this
Machine Learning Engineer | Awarded Inventor | Content Creating Specialist | MInd Mapping Proffessional | AI Enthusiast
آقا بیاید یه حالی به دیتاست فارسیش بدیم اگه خواستید می تونید از این لینک ثبت نام کنید، وارد شید و از فیلترها زبان فارسی رو انتخاب کنید https://lnkd.in/gaBQzyyt
3 days left to validate translations in your language an collaborate toward making multilingual benchmarks more equitable. Thanks to the 154 people who have contributed, and particularly to the 40 who have done over 100 validations! 🏁 8/39 languages complete. 🎊 Vietnamese, Portuguese, & Amharic are all over 1/2 way done! 🏃♂️ German, Indonesian, Sinhala, Turkish, Nepali, Swedish, and others have progress, but will need more help to cross the finish line before our Friday deadline! Join us: https://lnkd.in/gPuK57uW Track the progress: https://lnkd.in/gCkzNBxb
-
-
Cohere For AI reposted this
I am super happy to have contributed to this project, so we can have a community-built multilingual benchmark for LLMs. The Cohere For AI team told us what their vision was and I took my experience in data annotation to help them build a dataset in Argilla that would make the annotation process fast and easy for anyone who wants to contribute. 🤗 Check it out and help your language make it to the benchmark by checking a few translations! 👇
🌐 Contribute to build a truly multilingual benchmark for LLMs in collab with Cohere For AI. Help review some MMLU translations in your language! 🔥 The progress - More than 22K contributions, ~150 contributors - 8 languages are completed (Russian, Hindi, Telugu, Arabic, Spanish, Korean, French, Ukrainian) If you have 10 min, join this Hugging Face Argilla Space and start reviewing translations in your language: https://lnkd.in/dTyHaPEF Vietnamese, Portuguese, Amharic, German, and Indonesian are almost completed, help complete them so they can be included in the benchmark. There are many other languages that need contributions too. Watch the progress for your language: https://lnkd.in/dqzg6KQq
-
-
Cohere For AI reposted this
Join a Collaborative Effort to Develop Inclusive AI Solutions Across Languages. I'm currently participating in an open science initiative by Cohere For AI aimed at enhancing Multilingual Machine Learning Updates (MMLU), by translating it into various languages. Now at the final 2 days of the translation checking days, your expertise is crucial, especially as some languages have low coverage. I've been the only one so far reviewing translations into Swedish so far. So Swedish has only a bit over 100 checked and corrected translations. This effort is vital because, as a recent New York Times article highlights (https://lnkd.in/e7pxwW9e), the majority of AI models are predominantly English-trained, sidelining many global voices and potentially exacerbating technological inequities. Your skills can make a real difference. Whether fluent in another language or simply passionate about making AI more inclusive, your help is needed: 👉 Join the effort here: https://lnkd.in/e4GUY427 🌍 See the progress of your native language: https://lnkd.in/egTqfWPx Together, we can ensure AI serves everyone effectively. Let's drive change to make AI not only universally accessible but also high-performing!
-
This article by Alison Snyder for Axios examines the potential and consequences of using synthetic data for model training. Great to see our recent work by Luísa Shimabucoro, Sebastian Ruder, Julia Kreutzer, Marzieh Fadaee, and Sara Hooker on AI-generated data cited here. 📰 https://lnkd.in/dQ_fZzJH 📜 https://lnkd.in/e2TZnK2z
This is AI's brain on AI
axios.com
-
🎧Take C4AI on the road! We currently have 18 episodes of our podcast, *Exploring the unknown, together*, with more on the way as we sit down to chat with leading ML researchers and policymakers. This page will link you to your favorite podcast player https://lnkd.in/gWmVdu2b
Exploring the unknown, together.
buzzsprout.com