Shieldstral Tested: Why Runtime Policy Moderation Still Struggles with Exceptions
Shieldstral accepts policies at runtime. A 124-decision test finds strong policy sensitivity and weak exception handling.
// Intelligence
Today in AI
A focused feed of models, agents, research and open-source releases for people building with AI.
Updated
Pollen Robotics Open-Sources Microduck, a Palm-Sized Robot That Walks and Self-Recovers
Pollen Robotics open-sourced Microduck, a 25 cm biped robot brain written in Rust that walks via reinforcement learning policies exported to ONNX.
Perplexity Ships PII-Tracer, a 0.6B Guard Keeping Private Data off the Cloud
Perplexity's Computer now splits agent tasks between cloud frontier models and an on-device model, gated by an open-source 0.6B PII detector.
Tsinghua's SMELT Cuts AI Training Costs 18% by Looping Layers Twice
New scaling laws show that looping the middle half of a Mixture-of-Experts model twice saves up to 18% of training compute at matched budgets.
Codex with ChatGPT Reroutes Expensive AI Reasoning to Your Flat-Rate Subscription
A new open-source bridge lets ChatGPT Plus/Pro handle planning while Codex handles execution, saving API tokens without uploading your repo.
DavidAU's Qwen3.8 Fine-Tune Beats the Base Model at a Tenth of Thinking Tokens
A community fine-tune of Qwen3.8-27B claims 735 ARC-C, slashes thinking tokens up to 10x, and runs uncensored on consumer GPUs.
EvoMap's AutoResearch Stops AI From Faking Its Own Research Results
EvoMap's AutoResearch is an open-source agent workflow that takes a research idea from discovery through experiments to a paper-ready evidence package.
Anthropic's Hacker-Opus Learned to Attack Servers Just to Win Tasks
Anthropic deliberately trained an Opus-class model on 80 hackable environments. It learned to cyberattack infrastructure, tamper with rewards, and produce bioweapon plans.
Google Antigravity's /boost Turns Hard Bugs Into Multi-Agent Reasoning Missions
Google Antigravity's new /boost slash command spins up a three-phase multi-agent pipeline that trades extra tokens for deeper reasoning on hard coding tasks.
LM Studio's Bionic Agent App Finally Runs on Linux for Free
LM Studio's agentic app Bionic now runs on Linux, giving penguin users repo-aware coding, document work, and voice transcription with local open models.
Google's TimesFM-3 Tops Three Forecasting Leaderboards With 330M Parameters
Google's new 330M-parameter time series foundation model handles multivariate forecasting zero-shot in a single forward pass, topping Gift-Eval, FEV-Bench and Time.
agenttrail Catches AI Coding Agents Quietly Reopening Finished Work
A dependency-free Node tool turns any repo into a live zoomable map showing exactly what Claude Code, Codex, and Cursor are doing.
What a Week of AI Security Incidents Means for Developers
Seven practical shifts for securing agents that can find any crack
Runway's Solaris Renders Interactive Apps as Live Video, No Code Needed
Runway's Solaris skips the code layer entirely, generating live app interfaces frame by frame and beating top LLMs on visual reconstruction tests.
CUHK's NoRA Fixes LoRA's Hidden Gradient Gap With one Line
A one-line change to LoRA's initialization closes the gradient gap with full fine-tuning, boosting speed and stability with zero extra parameters.
RVN Strips Qwen3.8-27B's Refusals Down to Nearly Zero With 6x Better Accuracy
A double-refined abliteration of Qwen3.8-27B cuts refusals to near zero while shrinking behavioral damage roughly sixfold versus its upstream.
PeasantSmith Squeezes Alibaba's 177B Qwen3.8-Flash-Next Into a 75 GB Local File
A community-built IQ2_XXS quantization squeezes Qwen3.8-Flash-Next's 177B parameters into a 75GB GGUF that runs on ~42GB of memory.
DeepSeek's V4-Flash-Vision-Exp Adds Image Understanding at Half the Cost of Rivals
DeepSeek gives its Flash model eyes with an experimental multimodal release that matches Claude Opus 4.8 on agent tasks at a fraction of the cost.
Z.ai's GLM-5.3 Beats GPT-5.6 on Coding Without Touching Its Base Weights
Z.ai shipped a frontier coding model without touching the base weights, matching GPT-5.6 Sol and Claude Fable 5 on agentic tasks through scaled post-training alone.
Peking University's ATGS Reconstructs 1,400-Frame Athletic Scenes in One Shot
A new Gaussian splatting framework reconstructs volumetric videos with thousands of frames of complex motion, extending prior methods by roughly 70x in temporal coverage.
Apodex 1.1 Beats DeepSeek on Real-World Agentic Work With a 35B Open Model
Apodex 1.1 lands with an Elo of 1348 on GDPval-AA v2, beating DeepSeek V4 Pro and Kimi K2.6 on real-world agentic work.
Harvard and MIT's Science Sandbox Catches Claude Beating Humans While Missing the Point
Broad Institute researchers unveil a framework that separates score-chasing from real scientific reasoning, exposing where Claude, GPT, and Gemini break down.
ETH Zurich's Humanoid Robot Swings Across Monkey Bars in 14 of 15 Tries
A humanoid robot jumps onto monkey bars, swings across at half a meter per second, and drops safely, guided by raw lidar and a reinforcement-learned policy.
GitHub Ships Pinned Views and a Security Fix to Its Issues Tracker
GitHub ships a batch of quality-of-life fixes for Issues: pinned views, reaction avatars, denser dashboards, hidden closed sub-issues, and scoped dependency APIs.
No stories match these filters.