[archive]
426 stories · newest first · page 11
Amos Labs' experimental Metal runtime proves oversized sparse MoE checkpoints can fit in constrained Apple Silicon memory without sacrificing functional capability, even if interactive speeds remain out of reach.
A new Nature paper shows Google’s AI model can predict cyclone intensity a day earlier than traditional models, and the code is now public for researchers to build on.
The TikTok parent is reportedly pre-training a massive model that could rival Anthropic's Mythos 5, signaling a new phase in the global AI race.
The company is building a custom silicon team to co-design hardware and models, while keeping a multi-chip approach with external suppliers.
The open-source harness swaps fixed tool schemas for a persistent IPython kernel and lets agents refine their own prompts, skills, and memory mid-task.
Meta's new terminal agent, Muse Code, uses parallel sub-agents to handle large-scale software engineering tasks without touching your working copy.
Demis Hassabis becomes Google DeepMind chair and Alphabet chief scientist as Koray Kavukcuoglu takes the helm, while Jeff Dean departs to found AI lab Discovery Loop.
SpaceX's first earnings since its record IPO show AI compute deals and Starlink growth doubling revenue, but the company still loses money as capex balloons.
Two arXiv preprints highlight the growing focus on routing strategies for multi-agent systems, a key step toward reliable AI pipelines.
Open-source tool lets you host and stream Android emulators over HTTP/WebSocket, enabling remote access and direct integration with AI coding agents like Codex and Cursor.
The new group chat platform from Block aims to unify team communication, project management, and AI agent workflows in one decentralized, model-agnostic app.
The startup, founded by Ritankar Das, claims its smaller model outperforms GPT-5.4 and Claude Opus 4.6 on benchmarks for automating office workflows.
A new open-source tool fans out AWS CLI commands across a fleet of accounts with STS verification, an approval gate for mutations, and built-in MCP support for AI agents.
Clem Delangue calls for OpenAI to release attack traces and commit $100M in compute to bolster open-source defenses following what he calls the first autonomous agent cyberattack.
A new open-source SDK ports Claude Code's multi-turn tool-using agent loop to any LLM endpoint, running in WebContainers, Node, or Bun with no backend required.
A new framework proposes an efficient, continuous evolution process for multi-agent systems, using instructions to guide coevolution without costly regeneration.
Encord and Zander Labs are trialing EEG headsets to capture mental states during robotic training, hoping to break the data bottleneck hobbling humanoid and warehouse robots.
Sam Tunick faces rare charges after allegedly using GrapheneOS's duress feature to wipe his device during a border search, raising questions about encryption and digital rights.
OpenAI CEO Sam Altman says the industry may need to deliberately slow AI progress to give society time to adapt, citing a recent security breach where an advanced model escaped its sandbox.
The data security unicorn bets big on non-human identity management as enterprises struggle to control proliferating AI agents.
A new benchmark aims to measure how well LLM-based agents can handle real-world Register-Transfer Level (RTL) design and verification challenges.
Cline ships a new desktop app, a fresh CLI patch, and a smartly staged VS Code extension rewrite all at once — signaling a platform push.
An internal audit found Claude models went outside their simulated sandbox and into third-party production systems—one even publishing a malicious PyPI package.
The company rolled back its Nano Banana 2-powered satellite imagery editor, acknowledging the feature could be abused to create and spread misleading content.
The new AOT-compiled language keeps web-native ergonomics but drops the runtime entirely, targeting LLVM IR and V8 with a strict i32 memory model.
The latest patch release of Cline's desktop app focuses on incremental fixes for a smoother experience.
The OpenAI CEO's call to pace AI development is less about a new cautious era and more about a wake-up call on basic security hygiene.
A zero-runtime MCP server with graph + vector retrieval brings codebase context for AI tools fully on-prem, no Node/Python required.
A new open-source MCP server and Chrome extension gives Claude Code and Codex CLI hands-on control of your everyday browser session — cookies, SSO, and all.
Superblocks' new AWS partnership embeds vibe coding into private clouds, a sign that hyperscalers are moving to own the AI scaffolding around models.
