BlackPrism
TechCrunch AI
Manus co-founders Xiao Hong and Ji Yichao have been barred from leaving China after the AI startup's $2 billion sale to Meta, as Beijing investigates whether the deal violated foreign investment rules.
geopoliticscompanymarketArs Technica AI
Disney cancelled its planned $1 billion partnership with OpenAI following OpenAI's announcement to shut down the Sora video-generating app, with reports indicating no money ever changed hands in the deal.
companyproductsArs Technica AI
Google Research's TurboQuant compression algorithm reduces LLM memory usage by 6x without sacrificing quality, achieving 8x faster attention score computation on Nvidia H100 accelerators.
modelinfrastructureresearchTechCrunch AI
Google Research announced TurboQuant, an AI memory compression algorithm that could reduce inference memory by "at least 6x," drawing internet comparisons to HBO's fictional "Pied Piper" compression technology.
researchinfrastructuremodelTechCrunch AI
Anthropic's fifth economic impact report finds no evidence of AI job displacement yet, but reveals a growing skills gap as early Claude adopters gain advantages, with CEO Dario Amodei predicting AI could eliminate half of entry-level white-collar jobs within five years.
labormarketTechCrunch AI
Deccan AI raised $25 million in Series A funding led by A91 Partners to expand its AI post-training data services using an India-based workforce of over 1 million contributors serving customers including Google DeepMind and Snowflake.
venturecompanymarketNVIDIA Blog
NVIDIA CEO Jensen Huang unveiled the Vera Rubin computing platform at GTC 2026 in San Jose, projecting $1 trillion in revenue through 2027 and launching DLSS 5 for GeForce users.
productsinfrastructurechipsNVIDIA Blog
NVIDIA CEO Jensen Huang says AI's future requires both proprietary and open models, with NVIDIA launching the Nemotron Coalition and partnering with Mistral AI on open foundation models.
modelcompanyinfrastructuremarketHugging Face Blog
IBM Research released Mellea 0.4.0 and three Granite Libraries for building structured, verifiable AI workflows on IBM Granite models.
productsmodelresearchagentsHugging Face Blog
Hugging Face's Spring 2026 report shows 13 million users and 2 million public models, with China now surpassing the U.S. in downloads at 41% and NVIDIA emerging as the top Big Tech contributor.
marketinfrastructurecompanyNVIDIA Blog
NVIDIA OpenShell is an open source runtime that sandboxes autonomous AI agents and enforces security policies at the infrastructure layer, preventing compromised agents from overriding policies or leaking credentials.
agentssafetyinfrastructureNVIDIA Blog
Emerald AI collaborated with NVIDIA, National Grid, and EPRI to demonstrate how power-flexible AI factories can autonomously reduce energy consumption during grid stress, stabilizing power networks while enabling faster data center grid connections.
infrastructureresearchNVIDIA Blog
NVIDIA has donated its Dynamic Resource Allocation Driver for GPUs to the Cloud Native Computing Foundation, moving GPU orchestration for Kubernetes from vendor-governed to community ownership.
infrastructureHugging Face Blog
Ulysses Sequence Parallelism enables training LLMs with million-token contexts by distributing attention computation across multiple GPUs through attention head parallelism. The technique is now integrated across the Hugging Face ecosystem.
infrastructureresearchHugging Face Blog
Hugging Face researchers surveyed 16 open-source RL libraries to identify best practices for asynchronous training, finding that disaggregating inference and training across separate GPU pools can dramatically reduce GPU idle time.
infrastructureresearchHugging Face Blog
H Company released Holotron-12B, a multimodal computer-use agent model built on NVIDIA's Nemotron architecture that achieves 2x higher inference throughput than its predecessor while excelling at web navigation and UI interaction tasks.
modelagentsHugging Face Blog
NVIDIA published a tutorial showing how to fine-tune embedding models for domain-specific RAG systems in under a day using synthetic data generation, achieving up to 26% improvement in recall on real-world datasets.
modelinfrastructureDeepMind Blog
Google DeepMind released an upgraded Gemini 2.5 Flash Native Audio model with improved function calling and instruction following for live voice agents. Live speech translation supporting 70+ languages is now rolling out in the Google Translate app beta on Android.
productsmodelHugging Face Blog
ServiceNow-AI introduces EVA, the first end-to-end framework that jointly evaluates voice agents on task accuracy and conversational experience using realistic bot-to-bot audio simulations.
researchagentsDeepMind Blog
Google DeepMind released Gemma Scope 2, an open suite of interpretability tools covering the entire Gemma 3 model family (270M to 27B parameters) to help researchers understand and trace risks in language model behavior.
safetyresearch