All stories
Perplexity Gives Users More Control Over How Computer Solves Tasks
Perplexity is giving users new effort controls for its Computer AI agent, letting them balance reasoning depth, coordination, speed and computing costs. Web access arrives first.
OpenAI’s GPT-Live-1 Raises the Stakes in Voice-Agent Economics
OpenAI’s GPT-Live-1 separates real-time conversation from complex reasoning, promising faster, cheaper voice agents while raising questions about benchmark claims, routing complexity and whether modular AI can improve business outcomes.
Gemini 3.8 Puts Real-Time Voice Reasoning to the Test
Google DeepMind’s Gemini 3.8 Live models bring real-time voice, vision and background reasoning to assistants, but their success may depend on timing, transparency and knowing when not to interrupt.
CHIVE’s Warning: Seeing Inside a Model Is Not the Same as Explaining It
Anthropic’s CHIVE study finds activation-reading tools did not outperform transcripts at predicting behavior changes, underscoring why interpretable features are clues, not proof, of causation inside AI models.
Apple’s Reported M8 Ultra Server Could Bring Macs Into the Data Center
Apple is reportedly exploring a 2029 enterprise AI server built around M8 Ultra chips, potentially bringing energy-efficient Apple silicon, Nvidia networking and Macs into data centers.
Starlink V5 Production Signals a New Hardware Push for SpaceX
SpaceX says its next-generation Starlink V5 terminal has entered production, but specifications, pricing, availability and intended customer segments remain unclear as the company targets broader satellite broadband growth.
Salesforce’s DarwinX bets agent advantage on harnesses, not model weights
Salesforce’s DarwinX reportedly raised a browser agent’s WebArena-Infinity score from 43.5% to 93% without changing the model, highlighting harnesses and evaluation as enterprise AI battlegrounds.
Anthropic folds Cowork into Claude and brings documents and slides inside chat
Anthropic is folding its Cowork agent into Claude, adding native document and presentation tools as it turns the chatbot into a broader workplace platform for modern work.
OpenAI’s Misalignment Framework Tests How Transparent AI Safety Can Become
OpenAI’s new misalignment framework promises criteria and timelines for disclosing problematic model behavior, but key questions about definitions, evidence, severity and confidential investigations remain unresolved.