AI & Machine Learning
Image / techcrunch.com
Featured
Meta releases Muse Wonder models for coding and memory
Wonder is available through Meta’s model API, while Wonder Full is limited to Muse Code subscribers.
Failed startups’ archives are becoming AI training data
Liquid AI Puts Structured Decisions on Small Devices
Microsoft’s StudentSim Tests AI Tutor Guidance on Simulated Learners
ReplaySSM Cuts Memory Work in Hybrid AI Decoding
Latest AI & Machine Learning
Nadella wants an emergency brake for AI systems
The Microsoft CEO proposes separating models from their control software, recording their actions, and giving authorized people a shutdown switch.
Why AI Refusals Can Fail
MIT Technology Review’s newsletter preview describes safety tradeoffs when useful skills overlap with dangerous ones.
Jev turns AI output into decisions, not text
TypeSafe AI says its model targets software automation, while its rapid enterprise adoption remains a company-reported claim.
Anthropic Test Sent False Homicide Tip to Philadelphia Police
A website-interaction test reached a public police tip line, but the available account does not explain how or why.
NVIDIA’s cuOpt cuts communication in giant planning problems
The multi-GPU solver is a developer tool for supply-chain and energy models, with gains that depend on size and sparsity.
How AI Prepares a Robot for Simulation
NVIDIA’s developer walkthrough uses a YuMi robot, frontier AI, and Isaac Sim to turn CAD files into a testable digital model.
Pollo AI turns creative ideas into image and video ads
OpenAI announced the tool as a way to create detailed images and cinematic video advertising with three named models.
AVEVA Puts Limits Around Industrial AI
The company reports pilots that adjusted equipment settings, while keeping operating boundaries and human responsibility central.
Agent Lightning Trains Agents Through Their Real Harnesses
Microsoft Research Asia’s open-source framework connects deployed agent code to reinforcement learning without rebuilding the agent loop.
Microsoft Unveils PCs for Local AI Development
The Surface Laptop Ultra starts at $2,600, while a matching developer workstation starts at $6,000.
OpenAI and Ironclad target AI agents for contract work
The companies are training and evaluating agents, but have not described tasks, scores, or customer availability.
OpenAI Shares Lean Proofs From Model-Assisted Math
The release connects a frontier AI model to formal mathematics, giving researchers material they may be able to inspect and check.
Musubi Turns Written Rules Into Moderation Labels
PolicyLM-1.7B is an open-weight announcement, not a verified production system.
Why People Use AI While Feeling Worse About It
Adoption is rising across countries, while surveys cited by Technology Review show growing nervousness and opposition.
Enterprise AI Needs Connected Workflows, Not Just Better Models
A custom MIT Technology Review Insights report says companies must redesign processes, data access, and controls before AI agents can work across the business.
Teaching Speech AI Saudi Arabic Without Losing Key Skills
NVIDIA’s workflow combines selective data cleanup, replay examples, and decoder settings for Najdi and Hijazi speech.
AutoSynthData Turns Agent Failures Into Training Tasks
ServiceNow reports improved benchmark results after targeting synthetic tasks at an agent’s hardest enterprise workflows.
DIN Deploy Connects ONNX Models to Local C++ Apps
NVIDIA’s open-source samples show how exported models can run with hardware acceleration on Windows and Linux.
Briefing
NVIDIA Libraries Let GPUs Reach Object Storage Directly
NVIDIA Designs Coding Agents Around Reversible, Testable Changes
NVIDIA VSS 3.3 Builds Video Agents with Less Manual Work
Reported Safety Problems Halt OpenAI’s Astra 6.1 Launch
OpenAI’s Safety Plan Covers Three Frontier-AI Areas
