News
Short posts prepared from approved sources with clear links back to the original articles.
RSS feedLatent Space interviewed Radical AI’s Joseph Krause on self-driving labs and why materials discovery depends on lab execution as much as AI models.
Data and agent context-intelligence features meant to help AI agents safely access scattered business context at scale.
OpenAI and Molecule.one showed a near-autonomous AI chemist using GPT-5.4 to improve a difficult drug-making reaction in medicinal chemistry.
GitHub introduced a Copilot desktop app designed as a control center for parallel agentic coding workflows while keeping engineers in charge.
A new arXiv paper introduces MemTrace, a benchmark that evaluates agent memory by knowledge point across age, question type, and evidence conditions.
Zhipu AI released GLM-5.2 under the MIT license, claiming a 1-million-token context and strong results on long-running coding benchmarks.
Anthropic became the first AI startup to join the Frontier carbon-removal coalition, a move that comes as AI energy use faces growing scrutiny.
A new Pew Research poll found that AI chatbot use has grown sharply, while 63% of Americans say the technology is advancing too quickly.
Anthropic was forced to restrict access to Fable 5 and Mythos 5 after abrupt export-rule pressure, highlighting uncertainty around AI model access controls.
Runpod published a guide for creating animated GIFs from still images using its Stable Diffusion template, inpainting, and frame stitching.
Autonomous agents in Amazon Quick, adding continuous task execution, an activity feed, and cross-source business insight discovery.
Workers in Shenzhen are training humanoid robots by controlling them with VR-style body rigs.
Google’s $100 Home Speaker is available for preorder after months of delay, with Gemini positioned as the main upgrade over audio quality.
The fight over Anthropic’s Claude Fable 5 and Mythos 5 reflects a broader reality: models with stronger hacking capabilities are coming regardless of policy friction.
Microsoft is reportedly moving Copilot Cowork toward usage-based pricing and considering a fine-tuned DeepSeek model as a cheaper option.
New reporting says Apple is still targeting 2027 for AirPods with cameras, part of a broader push toward AI-capable wearable hardware.
A new technical guide walks through using xFormers with packed sequences, grouped-query attention, ALiBi, SwiGLU, and causal attention for efficient Transformer models.
AWS introduced container image caching for SageMaker AI inference, saying it can cut end-to-end latency by up to 2x during scale-out events for generative AI models.
NVIDIA made XR AI available in public beta, giving developers a framework for building multimodal AI agents for AR glasses and XR devices.
New market data cited by TechCrunch says ChatGPT remains the largest AI assistant but has fallen below 50% share for the first time as Gemini and Claude grow.
LangChain detailed how Monte Carlo built an AI Troubleshooting Agent with LangGraph and used LangSmith to debug workflows for data teams.
Google DeepMind and the UK government are building an AI-powered planning prototype aimed at speeding up housing decisions.
Factory used LangSmith to debug issues and automate feedback loops, doubling iteration speed on its AI product workflows.
AWS introduced the InvokeGuardrailChecks API for Amazon Bedrock Guardrails, letting developers apply individual safeguards inside multi-turn agent workflows.
Google Research described an Earth AI approach that turns environmental imagery and data into planning tools for nature restoration work.
HPE and NVIDIA expanded their AI Factory offering with components aimed at moving agentic AI from proofs of concept into enterprise production.
A Berlin court ruled that Google’s AI-generated summaries are a new search result format, partly contrasting with a Munich decision on liability for false AI responses.
AWS published a guide to using P-EAGLE speculative decoding on SageMaker AI to speed real-time generative AI inference.
A WordPress VIP survey found that 60% of US consumers view “AI” in brand messaging negatively, despite companies treating AI search as a growing referral channel.
A new arXiv paper proposes DivInit, a training-free method that improves parallel agentic search by seeding rollouts with more diverse initial queries.
The Pentagon says it is using generative AI to help write reports required by Congress and claims 1.5 million personnel are using the tools.
A new arXiv paper introduces PromptMN, a pseudo-prompting language that annotates natural language prompts with compact typed directives for roles, goals, constraints, and outputs.
Agent engineering combines product thinking, engineering, and data science to build and iterate reliable LLM systems.
A new arXiv paper introduces MemSlides, a memory-driven agent framework for personalized presentation generation and multi-turn local revision.
A new arXiv paper proposes an LLM-based agent that iteratively creates and tests query-rewriting rules to improve BM25 legal case retrieval without parameter training.
A new arXiv paper proposes structural uncertainty, a framework for assessing whether LLMs can consistently rank competing reasoning paths, not just final answers.
Coherent broke ground on an expanded Texas manufacturing facility for lasers, optical components, and compound semiconductors used in AI infrastructure.
Google released Android 17 and Wear OS 7 with new multitasking, safety, and smartwatch features, alongside a Pixel Drop that brings newer AI models to devices.
LangChain introduced Open SWE, a cloud-hosted coding agent that can plan, code, test, and open pull requests for GitHub tasks.
A new arXiv paper introduces Dr-DCI, a retriever-steered direct corpus interaction framework for agents searching and verifying evidence across large corpora.