AI for climate and scientific simulators
Learned emulators accelerated climate and materials simulations, complementing classic HPC.
Sorted by date · America/Guayaquil timezone
Latest signals
Page 31 / 38 · 447 articles
Autonomous coding agents (2025): from autocomplete to PRs
Autonomous-engineer products plan issues, edit repos, and open pull requests with variable human oversight.
World models and video understanding (2025)
World-model research sought dynamics prediction for planning—beyond pretty pixel generation.
Llama 3 and 4: Meta sustains the open ecosystem
New Llama generations improved multimodality and agents while keeping downloadable weights as industrial strategy.
The 2024–2025 open-weight frontier race
Meta, Mistral, DeepSeek, and others brought public weights near closed-API performance, redefining “frontier.”
Blackwell, TPUs, and custom ASICs: the 2025 silicon war
NVIDIA Blackwell, new Google TPUs, and hyperscaler ASICs defined who trains and infers at what cost.
Synthetic data: training models with models
Labs generated synthetic corpora for reasoning and code—with gains and feedback-collapse risks.
o3 and DeepSeek-R1: the test-time reasoning race
Models optimized to think longer at inference defined 2025, with leaps in math contests and coding agents.
DeepSeek: Chinese open models that shook cost assumptions
DeepSeek released competitive reasoning models with reportedly low training costs, altering market expectations.
Foundation models for robots: from factory to home
Vision-language-action policies and teleop data advanced generalist manipulation—still far from perfect home robots.
Small language models: intelligence at the edge
Fine-tuned, quantized 1–10B models brought useful assistants to phones and PCs without a full cloud round trip.
Computer use: agents that operate graphical interfaces
Labs showed models that click, type, and navigate UIs like humans, extending automation without APIs.