Google's ERA AI Tops CDC Flu Forecast Challenge Beating 39 Teams
Google's AI-generated forecasting model beat 38 other submissions in the CDC's FluSight evaluation, built by an LLM plus tree-search system now in trusted tester preview.
20 stories
Google's AI-generated forecasting model beat 38 other submissions in the CDC's FluSight evaluation, built by an LLM plus tree-search system now in trusted tester preview.
OpenAI released mcp-extensions, an open-source SDK that lets Model Context Protocol plugins render as native ChatGPT surfaces like sidebar apps, file viewers, and composer mentions.
Google's new flagship Gemini 4 Argon claims the top slot on the Vals Index at 68.9%, beating Sonnet 5.5, Opus 5, and GPT-6 while using far fewer tokens.
Google DeepMind's new frontier model pushes output to 1M tokens, targets long-horizon coding, enterprise workflows, and autonomous cyber defense through the Fairwind Program.
An ambitious MiniMax H3 VAE project failed its original goal but shipped a working 2X decoder and reference detail enhancer for ComfyUI.
Inception's new Mercury Decide is a structured decision model that returns typed answers with calibrated probabilities at 14 decisions per second, free on OpenRouter.
Perplexity's new 9B contextual embedding model distills relevance from a compression teacher, beating Voyage on turbopuffer's private context-bench with 8x less storage.
Runway's Solaris generates application interfaces pixel by pixel in real time, skipping code entirely and letting a world model render each frame as you click.
An open-source Claude skill brings disciplined logo design to AI coding agents, with SVG audits, a 1,400 logo reference library and testing tools.
An open-source Java platform stitches scripts, storyboards, image and video generation, and dubbing into one self-hosted workflow for AI dramas and comics.
Runway is bringing its video pretraining stack to robotics with an open-weight policy model that generalizes across robot bodies and environments.
A new benchmark shows frontier agents can match human predictive accuracy on scientific problems but still fail to produce real understanding.
A 260M-parameter text-to-image model beats one 6.5x larger by running the same transformer blocks multiple times per denoising step.
Pangram Labs trained 800 language models to measure how AI-generated web text warps pretraining, and built a new scaling law to price each token.
NVIDIA's PivotOPD pinpoints the single early mistake that dooms most agent rollouts, then teaches the student to both avoid it and recover.
GitHub extends its HydraFusion research preview from the Copilot CLI into VS Code and the Copilot app, exposing multi-model routing to more developers.
Google is retiring Opal and folding its no-code workflow ideas into Gemini Skills, a slash-command system replacing Gems worldwide.
Runway just turned its internal ad machine into a product that generates, publishes, and iterates on Meta, Google, and TikTok creative autonomously.
Utopai Studios post-trained MiniMax H3 into a video model that debuts second on Artificial Analysis' text-to-video board, trailing only Alibaba's Wan 3.0.
Zed 1.22 lets the built-in agent spawn subagents on different models, adds automatic context compaction, and dynamically fetches OpenCode models without app updates.