Newsroom
AI news, from here and the world
Center announcements, research releases and what matters in AI for the region — in four languages.
Curated links — updated weekly by the editor
Models
Open-weight models are closing in on the frontier
The best open models now trail the frontier labs by only a few months — the gap that matters for a center that cannot afford frontier-scale training runs keeps shrinking.
Epoch AI
Policy
GAO: agencies should apply hard lessons to their next AI procurements
Data rights, requirements definition and testing keep sinking AI contracts — the checklist any public-sector AI center should read before its own procurements.
U.S. GAO
Data
Data quality beats model size on narrow tasks
A study of 400+ models shows redundant, low-diversity data stalls scaling — good news for teams whose advantage is owning data nobody else has.
ACL 2025
Speech
Speech recognition for low-resource languages is still where the field struggles
The same data-scarcity wall we hit with Karakalpak STT, documented across dozens of languages — with the community-driven collection strategies that actually help.
arXiv
Vision
ShieldGemma 2 takes vision-language safety classifiers out of the lab
An open, fine-tunable multimodal safety model shipped as a production component rather than a research demo — a template for scene-understanding safety work like ours.
Google DeepMind
Compute
Inference cost per token keeps falling
Prices for a fixed capability level drop 9×–900× per year — on-premise deployment for public institutions gets cheaper every quarter.
Epoch AI