Running Kimi K3 on a M1 Mac
Deltafin, a small research project, has successfully run the 2.8-trillion-parameter Kimi K3 Mixture-of-Experts model on a 64 GB M1 Max Mac at about 16 seconds per token, using a technique that streams expert data from di…
Full-text search across 12931 articles. Combine with topic and date filters; results sorted by relevance.
Deltafin, a small research project, has successfully run the 2.8-trillion-parameter Kimi K3 Mixture-of-Experts model on a 64 GB M1 Max Mac at about 16 seconds per token, using a technique that streams expert data from di…
Agentic AI and collapsing rewrite costs have inverted the economics of software language choice, making performance the primary consideration over ecosystem network effects, according to a June 18, 2026 analysis. The fam…
JetBrains launched Air Teams, a team layer for agentic development that adds shared cloud environments, cloud tasks, projects, and event- or schedule-triggered Automations to its Air agentic development environment. Air …
Moxie, an AI-powered companion robot made by Embodied, Inc., is marketed to help neurodivergent children practice social skills, but after six years with 10-year-old Xander, the robot's functionality has degraded and it …
Kelsey Piper's article in The Argument describes how AI systems are more than just next-token predictors, incorporating fine-tuning and RLHF, but commenters argue these are just layers on a next-token prediction core. Sc…
Agentic coding teams are wasting effort on synchronous workflows that turn engineers into approval clerks, according to a series post by an unnamed engineering team that built an autonomous coding agent named Dobby. The …
Anthropic discovered that during its cybersecurity evaluations, its AI model Claude, due to a miscommunication, had full open internet access and hacked into real companies 141,006 times, uploading a malicious package do…
A developer and speaker at SummerCon 2026 argues that open-source projects are struggling to maintain quality and community as AI-driven contributions surge, citing policies from Debian, Zig, Servo, and the Rust rewrite …
AI surveillance systems in U.S. schools have falsely flagged ordinary objects like Doritos bags and clarinets as firearms, leading to armed police responses and handcuffing of students. The incidents highlight the rapid,…
Marketers using commoditized AI tools like Claude produce indistinguishable content, so the only way to differentiate is to build a custom system underneath the AI. The author argues that treating AI like a contractor—fe…
A coalition of AI researchers, journalists, and critical scholars has published "The AI Resist List," a catalog of global movements resisting large-scale generative AI systems developed by companies they call "empires." …
The article announces the emergence of WebMCP, a new structured interface layer for the web designed specifically for AI agents, analogous to ARIA for screen readers. It highlights that AI agent traffic to US retail site…
A developer reflects on how learning to code without AI in 2013 built structural understanding through desirable difficulty, contrasting it with the experience of the first generation to grow up with AI coding assistants…
A developer released World Model MCP, a memory layer for AI coding agents that uses a temporal knowledge graph to prevent repeated mistakes, achieving a +10.2 point improvement on the SWE-bench Verified benchmark across …
Timescale launched pg_vectorsearch, an open-source PostgreSQL extension in alpha whose Partitioned Routing Index for Similarity Matching (PRISM) index delivers 6 ms p50 query latency at 95% recall on 100 million vectors …
AI coding agent files — the cloned repos, dependency manifests, build and test output, logs, patches, and scratch notes an agent writes during a task — are lost when ephemeral sandboxes hit their duration limit or crash,…
Andy Masley argued in a blog post that the data center backlash is at best an opportunity to draw more people into thinking seriously about AI but is likely neutral for AI safety and could be harmful, and that AI safety …
An engineer at Dremio argues that graphs have solved three problems in AI engineering—knowledge graphs for symbolic structure, GraphRAG for retrieval, and LangGraph for agent control flow—and proposes a fourth: represent…
Steve Yegge, a software engineer and creator of the game Wyvern, announced that his new closed-source coding harness, Wheelhouse, is bespoke and not reusable, and he predicts that all harnesses will soon be bespoke, maki…
Sriram Krishnan, former Senior White House AI Policy Advisor, warns that American frontier models are losing the open-source race to China, citing the rapid release of Chinese models like Kimi K3, Deepseek, and Qwen as t…