Build Long-context RAG from scratch: Nomic Embeddings + Mistral The context window of open source LLMs and embedding models has been relatively small vs proprietary models. But, methods to expand context window (RoPE, self-extend) are quickly changing this. Today, @nomic_ai has
RESEARCH
-
Exploring new failure modes in AI systems beyond current models
By
–
melty videos will always have a special place. But I want new and different failure modes.
-
National AI Research Resource NAIRR Launch Explained
By
–
This video aged well: Published in 2021, here we explained why we were calling for the creation of a national research cloud—now materialized as the National AI Research Resource or NAIRR.
-
Strong Untestable Claims Incompatible Scientific Epistemology
By
–
Making strong untestable claims is incompatible with a scientific epistemology
-
Is the World’s Mechanism a Living Organism?
By
–
What does it entail if the mechanism of the world is a living organism, and how do you test for that hypothesis?
-
Africa’s AI Indaba 2026 Comes to Senegal
By
–
So excited that we will be in going to Senegal . This year’s Indaba is all about multi-linguality, showcasing more research, and sharing the amazing stories of African AI . Please support our fundraising and work.
-
Stanford AI+Education Summit Addresses Ethical, Equitable AI Teaching
By
–
Today @Stanford
: 250+ educators, students, and technologists gather for the second year of the AI+Education Summit. AI has the potential to transform teaching and learning, but how can we build it in an ethical, equitable, and safe manner? @StanfordEd https://
stanford.io/42uzjbK -
AI Explores Human Brain Visual Processing with Natural Synthetic Images
By
–
Researchers at Weill Cornell Medicine are harnessing AI to explore the universe's most complex known object, the human brain. Using AI-selected natural and synthetic images, they delve into the visual processing areas, aiming to understand organization and minimize biases. We
-
Fine-tuning Mistral 7B with DPO dataset for improved chat performance
By
–
I fine-tuned Mistral7Bv0.2 locally with MLX using this DPO dataset by @argilla_io
, composed of 7k chat interactions distilled from Capybara. I haven't done a full evaluation yet, but for this size, it's really good, and it understood this famous question right away! Link -
New Zealand Broadcaster Uses Trustworthy AI for Māori Language
By
–
Learn how New Zealand broadcaster @TeHiku uses #trustworthyAI to preserve the indigenous language of the Māori people, known as te reo. Built using the NVIDIA NeMo toolkit, their automatic speech recognition model transcribes te reo with 92% accuracy.