Big Context Windows Are Not Memory: What I Learned When My Chats Kept Dying
A student's journey from a frustrating ChatGPT session to building an open-source framework for deciding what an LLM should actually remember.
Open Science Community
Cohere Labs Community Blog
Research notes, technical essays, and personal stories from the Cohere Labs Community
Research notes, stories, and ideas from people shaping AI together—transparent, collaborative, and community-led.
Latest posts
-
-
NILEAGI-SUB: An Evaluation Programme for Swahili Language Understanding
We built NILEAGI-SUB to measure Swahili understanding where people actually use it. Our first sector is education: 2,569 school questions, eight compact open models, and a ranking that size alone cannot explain.
-
Language, Decoded: What if you could code in your own language?
Quality data for under-resourced languages is hard to acquire. It is even harder to get code in these languages. Through research, it showed that introducing code in training a model has proved to be a well rounded data source to improve performance. How can we utilize non-English code to strengthen the performance of these multilingual models, especially where data scarcity exists? We started with the question:- what if the code you train a language model on was written in a language other than English?
-
From Reviewing Words to Building Africa's AI: Our Journey Inside the Cohere Labs Open Science Community
How a Luganda word-review task became a startup, a Gold Award, and a lesson about what really keeps gates closed for African builders, and what opens them.
-
Inside Tiny Aya: What an SAE Sees in a Model Built for 70+ Languages
We trained and released Sparse Autoencoders on all four Tiny Aya regional variants and asked what an SAE sees inside a model built for 70+ languages: a representational-density gap, a mostly-shared feature basis, and an identity-versus-quality null.