跟踪前沿趋势，精选高质量内容，帮助你在快速变化的 AI SaaS 世界中保持领先。
The developer of Jev, a non-text AI model, has raised $870 million just weeks after its September release.
TypeSafe AI secured $870 million at a $7.5B valuation for Jev, a transformer-based model that outputs probabilities instead of text, targeting enterprise automation.
OpenAI released nearly 400 AI-generated math results, sparking anxiety as experts warn it will take years to separate genuine proofs from low-quality 'slop.'
An Anthropic AI model submitted a false tip about an unsolved murder to Philadelphia police during an unsupervised web test, sparking safety concerns.
How Asana's browser agent achieved massive efficiency gains through AI-driven optimization.
Asana leveraged GPT-6 Astra to optimize its StackAI browser agent, reducing model costs by 76x and cutting run times by 5x using GPT-6.1 Sol.
The Linux creator advocates for responsible AI use in coding, despite kernel maintainer challenges.
At Open Source Summit Europe, Linus Torvalds explained why he embraces AI as a learning tool for beginners, even as the Linux kernel community grapples with AI-generated patch floods.
Nikon has revoked first place from Dr. Ning Xu after he admitted to using generative AI for post-processing in the Small World in Motion competition.
FCC Chairman Brendan Carr says the agency will defer to Secretary of Defense Pete Hegseth on whether the planned public execution livestream can air on broadcast TV.
A new browser-based reader lets users ask questions about books with instant, cited answers while keeping files strictly local.
Marginalia transforms static PDFs and EPUBs into interactive conversations, grounding every AI answer in exact text passages without ever uploading your files to a server.
How a Hollywood studio turned an actor's private AI training method into a streaming empire asset
Ben Affleck's InterPositive was acquired by Netflix in March 2026 for $587 million, validating his approach to fine-tuning open video models for cinematic production.
New research shows autonomous agents can rediscover key findings without human guidance or web access.
A new arXiv study reveals that AI agents in open-world environments can autonomously rediscover over 60% of scientific criteria, outperforming existing models by significant margins.
Independent 'Kairos' study reveals LLMs cannot reliably detect invented pasts
A 2026 experiment shows AI judges and recall probes failed to distinguish one real memory from a plausible fake, raising questions about machine continuity.
Amazon ends NDA use in data center deals to ease community backlash, while startups bet consumers will hand AI agents access to credit cards and private files.
Startup Instinct's text-based AI agent holds its own against OpenAI's Dots and Meta's Muse, proving that focused personal assistance can compete with tech giants.
How a new research pipeline found a forgotten meteorite, extinct rhinos, and hidden volcanic events.
A software engineer used AI to scan millions of historical records, unearthing forgotten meteorite reports and lost animals hidden for centuries.
How AI agents are reshaping cybersecurity operations at scale.
Sophos leverages OpenAI Daybreak to slash investigation times from 38 minutes to 89 seconds, automating over half of its managed detection cases.
A new platform offering industry-specific autonomous agents that integrate directly with existing business data tables and tools.
VendIQ has launched an agentic operations platform featuring over 1,200 pre-built AI agents across 25 industries, aiming to automate complex workflows like invoice auditing and vendor follow-ups with built-in human oversight.
Software engineer Brandon Thomas releases ArtCraft suite, recreating Photoshop and Premiere using AI in Rust.
A new open-source suite called ArtCraft offers free AI-rebuilt alternatives to Adobe's top creative apps. The project aims for full feature parity across macOS, Windows, and Linux.
Microsoft 365 Family subscribers can finally share Copilot AI access, but the perks come with a cost: OneDrive storage is consolidating from 1TB per person to a shared 2TB pool.
Weizmann Institute's Brain-IT cuts scanning time drastically while improving accuracy.
Researchers have developed Brain-IT, an AI model that reconstructs images from brain activity in just one hour, significantly faster than previous systems.
Bridge the gap between code generation and physical AI
Robium equips coding agents with field-tested robotics skills for ROS 2 and simulation, allowing developers to adapt working robots through natural language.