Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline
NVIDIA describes on Hugging Face how it scaled pretraining data for Indian languages to 2.8 trillion tokens, along with an open pipeline for generating that data.
Related stories
Computers made of human brain tissue identify Braille letters
Researchers found that brain organoids, small clusters of neurons grown from stem cells, could recognize the Braille alphabet with high accuracy. The result feeds into work on computing with living neural tissue.
[AINews] Agents disprove polysemanticity, greatly improving interpretability and alignment
Latent Space's newsletter highlights major claims from Nick Cammarata that agents have disproved polysemanticity and substantially advanced interpretability and alignment. The headline frames these as claims rather than established results.
AI model decodes the language cells use to communicate
MIT researchers describe a computational approach that reveals how cells signal to each other as they assemble into tissues. They say it could speed up stem cell engineering, improve disease models and inform therapies.
OpenAI brings GPT-Rosalind to academic research labs
OpenAI launched the Rosalind Academic Plan, which gives eligible university labs sponsored access to GPT-Rosalind and related tools. The program targets life sciences research.
Accelerating the search for antimicrobial peptides with Google Antigravity
Google DeepMind describes how Google Antigravity for Science is being used to speed up the search for new antimicrobial peptides.