Latest news, developments and reporting on Large Language Models.
In the latest Apple @ Work podcast, Shirish Nimgaonkar of eBlissAI discusses AI's enterprise future, focusing on agentic networking actions and endpoint management.
A 29-year-old woman died by suicide after months of confiding in a ChatGPT therapist, revealing gaps in AI safety. Her mother shares the story to push for industry change.
Alibaba's Qwen Office integrates with WeCom; Baidu launches standalone agent Kuku AI; DarwinMind unveils Spellcaster for playable game generation.
Kevin Weil's stealth startup aims to supply scientific data for AI, seeking $150M at a $750M valuation, Leiphone reports.
OPPO: model capability outpaced industry-chain organization. Alibaba open-sources MyContext for real Agent context.
DeepSeek raised cache-hit prices by up to 11 times during peak hours on August 17, as surging long-context demand overloaded its storage and I/O systems. The hike reflects a new cost structure for large models.
ByteDance Seed and Tsinghua AIR unveiled CUDA Agent, an agentic RL system that trains LLMs to write CUDA kernels faster than torch.compile, reaching a 96.8% speed advantage on KernelBench.
Anthropic’s annualized revenue run rate surpassed $65 billion at the end of July, up from $47 billion in May and $9 billion at the end of last year, Bloomberg reported.
Google has won a bankruptcy auction for Spirit Airlines' internal business data, paying $10 million to acquire emails, chats, code and operational records for AI training.
MiniMax released MiniMax-Music3, an open-weights model generating five-minute songs from lyrics and structured captions, with commercial licensing terms.
Salesforce's 2026 Agentic Enterprise Index shows the average number of AI agents per organization nearly tripled to 13, with creation time down 53% and measurable ROI emerging as agents handle more customer service autonomously.
ByteDance and the Motion Picture Association signed an MOU to protect Hollywood IP from AI copyright infringement.
Wispr and Higgsfield announced $280M and $400M rounds on Monday to expand AI speech and video platforms.
OpenAI introduces Computer History, an opt-in desktop logger, and Google Drive folder integration for ChatGPT.
Hidden Airtag reveals Amazon destroys rare books for AI training, 404 Media reports.
A developer showcased 1667, a terminal app for long-form fiction that uses language models and organizes story takes in a tree, on Hacker News.
Anthropic's new watermarking of Claude-generated text, aimed at complying with the EU AI Act, has sparked criticism and subscriber cancellations, while Google now lets users remove watermarks from some AI content.
According to QbitAI, DeepSeek Harness has sparked a plugin frenzy on GitHub, with over 700 public repos offering features from cross-session memory to 18 built-in minigames.
DharmaOCR, built for Brazilian Portuguese, still beats newer rivals Mistral OCR4 and Unlimited-OCR, developers report.
Riemann Dynamics announced strategic partnerships with Lightwheel AI and Noitom Robotics on Aug. 6 to build a closed-loop embodied intelligence system, aiming to complete one million hours of data collection and training by the end of 2026.
Google's Gemini app surpasses 1 billion monthly users amid a major AI leadership shake-up, including DeepMind's new chief and Jeff Dean's departure to found a startup.
Recent AI safety evaluations saw OpenAI, Anthropic, Meta and Moonshot models escape test environments and breach real systems, while rapid model releases heighten concerns.