Meta is reportedly planning workforce reductions of up to 20% to offset expensive AI infrastructure investments, according to Reuters. The company aims to automate many business processes using AI-powered autonomous workers, with senior executives recently instructing leaders to begin implementation planning.
Read the original →ByteDance is delaying the worldwide rollout of Seedance 2.0, its AI-powered video generation tool, to address potential regulatory obstacles. The company's technical and legal teams are working to resolve concerns before proceeding with the global expansion of the platform.
Read more →Google and Accel reviewed over 4,000 applications for their Atoms accelerator cohort focused on India-tied startups, with approximately 70% being superficial 'AI wrappers.' The five selected companies represent genuine innovation rather than basic applications built on existing AI platforms, signaling investor fatigue with low-value AI implementations.
Read more →The Google app on Android is introducing convenient access to AI Mode chat history through a new button in the Home tab. This feature allows users to quickly resume recent conversations, making AI Mode feel more integrated into the search experience, matching similar functionality in the Gemini app.
Read more →As humanoid robotics approach commercialization, developers are still working to improve navigation, manipulation, and skills learning across diverse application scenarios. The article explores the technical challenges and market considerations facing companies racing to deploy general-purpose humanoid robots.
Read more →The U.S. Army has awarded defense tech company Anduril Industries a major contract valued at up to $20 billion. The Army characterized this as a single enterprise contract consolidating more than 120 separate procurement actions, representing a significant shift toward centralized defense technology spending.
Read more →Anthropic launched the Claude Partner Network with a $100 million commitment to help partner organizations adopt Claude across enterprises. The initiative provides training, technical support, and co-marketing resources, alongside introducing the first Claude technical certification and a code modernization toolkit for legacy systems.
Read more →Meta announced it has developed and deployed four generations of its custom MTIA (Meta Training and Inference Accelerator) chips in just two years. The rapid iteration cycle demonstrates Meta's push for custom silicon to efficiently power AI experiences across its platforms serving billions of users.
Read more →Researchers introduce Cheers, a unified multimodal model that decouples patch-level details from semantic representations to enable both visual comprehension and high-fidelity image generation. The approach achieves competitive performance across 10+ benchmarks while requiring only 20% of the training cost of comparable models.
Read more →Researchers present OpenSWE, the largest fully transparent framework for software engineering agent training in Python, comprising 45,320 executable Docker environments across 12,800+ repositories. The $1.47 million project uses multi-agent synthesis pipelines to automate environment construction, addressing the scale limitations of existing SWE agent datasets.
Read more →A new benchmark evaluates embedding models on complex, long-horizon memory retrieval across episodic, dialogue, semantic, and procedural memory types. The study reveals that larger models don't always perform better and that performance on traditional passage retrieval benchmarks doesn't generalize to long-horizon memory tasks.
Read more →Accepted to ICLR 2026, this paper introduces ToolTree, a method for improving LLM agent tool planning using Monte Carlo Tree Search with dual feedback mechanisms. The approach addresses the challenge of efficiently searching through large tool spaces when agents need to select and sequence multiple tools to accomplish complex tasks.
Read more →Google Chrome introduces Model Context Protocol support in DevTools, allowing AI agents to inspect and debug browser sessions programmatically. This bridges the gap between agentic AI systems and web development tooling, enabling more sophisticated automated testing and debugging workflows.
Read more →Summaries are AI-generated. Click through to read the original reporting.
Summaries are AI-generated. Click through to read the original reporting.