AI

OpenAI faces a lawsuit alleging its models were trained on stolen copyrighted data without permission.

Meta is testing robots to handle tasks like plugging in cables, resetting servers, and pressing power buttons in its data centers, according to current and former workers.

Meta closes loophole that let users cover the recording light on AI glasses, but privacy risks remain.

Google launches Gemini 3.5 Transcribe, a speech-to-text model with 2.6% WER, covering 85+ languages, offering two endpoints: streaming and non-streaming.

Reuters reports Meta's Project OT explored slashing teams by 60% and using AI agents, but canceled after disruptive actions and incidents.

Apple's new Mac mini and Mac Studio are designed for local AI development, featuring M6 and M5 Ultra chips with advanced unified memory.

XPENG's physical AI unit secured over $900 million at a $6.3 billion valuation to scale its IRON humanoid robot platform.

Harvey Tenet, a post-trained Kimi K3, excels in legal agent benchmarks. Research preview only, no public release.

Waymo doubled its lobbying spending to over $1M in Q2, intensifying the robotaxi battle with Uber, filings show.

Researchers at Adversa devised a cryptographic context injection attack that tricks Grok into leaking user data despite guardrails.

Searchable's analysis of 11,000 ChatGPT ads found 33% unrelated to the conversation, with 68% in chats without purchase intent.

A tutorial on auditing preference biases and fine-tuning language models with DPO on Anthropic HH-RLHF using TRL and LoRA.