All Updates
Page 1820 of 1910
February 24, 2026
MPA warns ByteDance Seedance 2.0 infringement
US Motion Picture Association sent a cease-and-desist letter to ByteDance over Seedance 2.0 AI platform's alleged copyright infringement. It demands halting use of studio movies and TV shows for model training and safeguards against user-generated protected content. This escalates from an earlier February 12 industry protest.
Deploying Open-Source VLMs on Jetson
Hugging Face blog details deploying open-source Vision Language Models (VLMs) on NVIDIA Jetson edge devices. This guide enables efficient on-device inference for multimodal AI tasks. Developers can now run VLMs locally without cloud dependency.
Rethinking Single Extractor for LLM Pretraining
Apple ML challenges the use of single fixed HTML extractors for web-scale LLM pretraining datasets, citing suboptimal coverage of diverse web content. Different extractors produce substantially varying pages after filtering, despite similar performance on language tasks. The work explores better extraction practices for improved data utilization.
depyf Demystifies PyTorch Compiler
PyTorch 2.x compiler accelerates deep learning at Python bytecode level but remains opaque to researchers. depyf decompiles this bytecode into equivalent source code. It connects in-memory code objects to on-disk source files for better debugging.
Apple Analyzes CoT Trace Dynamics
Apple ML conducts in-depth analysis of Chain-of-Thought (CoT) traces from competition-level math questions. The study uncovers how specific parts of CoT contribute to final answers in LLMs. It aims to clarify the mechanisms driving CoT's reasoning success.
AMUSE Benchmark for Multi-Speaker AV Agents
Apple introduces AMUSE, a benchmark for multimodal LLMs in agentic multi-speaker dialogue scenarios. Models like GPT-4o and Qwen3-Omni excel in perception but falter in tracking speakers, roles, and events across audio-video streams. It targets applications like conversational assistants and meeting analytics.
February 23, 2026
Anthropic Staff Share Sale at $350B
Anthropic enables current and former employees to sell shares at a $350 billion valuation. This follows a recent $30 billion fundraising round.
Claude Code refactors COBOL, slams IBM stock
Anthropic's blog demonstrates Claude Code accelerating COBOL app refactoring. IBM's share price plunged 13% amid investor fears over AI displacing legacy maintenance. Ironically, IBM has touted AI for COBOL since 2013.
Samsung Brings Perplexity AI to Galaxy Phones
Samsung plans to integrate Perplexity AI into Galaxy smartphones. This daily roundup also notes Linux 7.0-rc1 kernel release and Firefox 115 ESR ending support for Windows 7/8 this month.
VLMs Scale Data Annotation for Physical AI
Bedrock Robotics partnered with AWS via the Physical AI Fellowship and Generative AI Innovation Center. They apply vision-language models to analyze construction videos, extract details, and generate labeled datasets at scale. This enhances data preparation for autonomous construction equipment.
Fate launches agentic AI dating app
Fate is the first agentic AI dating app that interviews users on hopes and dreams, then matches five potentials via language patterns—no swiping. Opinion piece decries it as commodifying profound emotions.
Anthropic accuses Chinese firms of Claude theft
Anthropic uncovered three Chinese AI firms—DeepSeek, Moonshot AI, MiniMax—using distillation to extract capabilities from its Claude chatbot. It calls this industrial-scale IP theft. OpenAI issued similar accusations last month.
SMEs Want AI but Lack IT Staff: Survey Reveals
A joint survey by Neos and the Solo IT Admin Association highlights SMEs' AI adoption struggles due to IT staffing shortages. Despite labor shortages driving AI interest, lack of skilled personnel hinders implementation. It explores solutions for accelerating AI use in small firms.
Gartner: 90% Routine Jobs Gone to AI by 2029
Gartner predicts 90% of workers focused solely on routine tasks will be replaced by AI by 2029. The report highlights traits of AI-vulnerable talent. It outlines responses needed for companies and employees.
Persona Selection Model for LLMs
The Persona Selection Model (PSM) proposes that LLMs learn to simulate diverse personas during pre-training, with post-training eliciting and refining a specific 'Assistant' persona. User interactions are framed as engaging with this character-like entity. Empirical evidence from behavior, generalization, and interpretability supports PSM, influencing AI development strategies like anthropomorphic reasoning.
Treasure Code: SaaS Built in 1 Hour via AI Governance
Treasure Data launched Treasure Code, an AI-native CLI for its CDP, built by one engineer in 60 minutes using Claude Code. It relies on pre-built governance for secure production deployment. A three-tier AI code review pipeline ensures quality without human coding.
Pentagon Disputes Anthropic AI Guardrails
The Pentagon summoned Anthropic's chief over a dispute on AI limits. The AI company demands guardrails during Defense Department contract talks. This highlights tensions in military AI deployment.
Amazon $12B Louisiana data center vows self-funding
Amazon announced a $12 billion data center project in Louisiana. The company pledges to fully fund its energy, water, and other infrastructure needs. This deal emphasizes transparency and accountability in big tech expansions.
Anthropic Accuses Chinese Labs of Claude Theft
Anthropic accuses DeepSeek, Moonshot AI, and MiniMax of using 24,000 fake accounts for 16 million Claude interactions to distill capabilities. This violates terms of service and highlights industrial-scale knowledge theft. The company calls for industry-wide action amid US-China AI tensions.
Japan Targets Top AI Adoption Globally
Digital Minister Matsumoto outlines Japan's strategy to become the world's most AI-friendly nation. Digital Agency focuses on AI, cybersecurity, and medical DX within economic security and growth plans. Emphasizes public-private collaboration.