Open-weight Kimi K2.7 model now available on GitHub Copilot, offering more choice and lower costs; initial rollout to select plans.
// curated from Hacker News with AI
Open-weight Kimi K2.7 model now available on GitHub Copilot, offering more choice and lower costs; initial rollout to select plans.
Local AI empowers users to own, run, and modify models on personal devices while enforcing laws against misuse and safeguarding privacy.
Senior SWE-Bench evaluates agents as senior engineers using realistic tasks involving feature development, bug investigation, and code quality assessments.
Fake news stories about Alabama newspapers are AI-generated hoaxes; real outlets remain unaffected, highlighting AI fakery's spread.
Training a single transformer layer can match or surpass full RL parameter updates, with key gains concentrated in middle layers.
A tool enabling any LLM like Claude to watch, analyze, and extract meaningful scenes, audio, and transcripts from videos locally.
Use strict planning, frequent review, and proactive oversight to harness AI for high-quality, secure coding in security-critical systems.
NSA attempts to weaken cryptographic standards by backing less secure MLKEM, facing opposition and urging public action before July 7, 2026.
CLI tool using embedding models to detect similar, non-exact code duplicates across large codebases for refactoring.
AI makes experienced developers feel faster, but actually slows them down on large codebases due to review overhead; perception is inverted.
Weird Al Yankovic refuses AI commercial for business software, citing discomfort with AI’s impact on art and creativity.
zkGolf is a competition to build cost-efficient, verified circuits in Lean, scoring based on circuit compactness and correctness.
Claude bug causes it to skip questions after 60s without response, leading to loading errors on Linux in VS Code.
Fable and 10 LLMs proposed refactoring a complex AI node; Fable ranked top for proposals; GPT models excel at evaluation.
Meta's AI development is slower than expected, with restructuring and investments not yet yielding anticipated benefits.
Open-source Valmis enables secure, containerized AI workflow automation with 100+ integrations, memory, and multi-agent management.
Meta plans to build cloud services, sell excess AI capacity, compete with top providers, and cut reliance on ad revenue amid AI race pressures.
Introduces QUALITY.md, an open standard for defining, evaluating, and improving project quality across domains using agent-agnostic tools.
‘Weird Al’ Yankovic declined AI commercial for a business app, citing discomfort with AI's impact on art and creativity.
Nvidia offers startups compute-for-profit deals, sharing future revenues in exchange for GPU access, fueling AI development.
Karp exposes AI labs stealing enterprise IP, charging for compute not value, risking industry trust and enterprise innovation.
AI note-takers blur the line between private and recorded conversation, reducing genuine connection and raising societal etiquette concerns.
AI essay contest on "Klara and the Sun," with a $1,000 prize, uses AI scoring and encourages thoughtful analysis of AI themes.
Leaked: Microsoft developing a lightweight Edge-based Windows 11 AI OS, details still under verification.
Traditional media's role as democracy's last defense is crucial amid rising AI-generated misinformation and declining trust in digital sources.
Users worry about losing control and opting out of AI features as big tech integrates AI into search, increasing usage but raising privacy concerns.
Alex Karp criticizes frontier AI models for failing to deliver meaningful outcomes.
Companies are limiting employee AI use due to unchecked costs, with spending soaring and access being cut back across multiple industries.
As AI floods the web, genuine content stalls while volume rises, risking epistemic heat death and a trust crisis.
BlastRadar instantly assesses production risk from code diffs, aiding quick decision-making for database and infrastructure changes.
New insights challenge traditional mean-field theory, potentially improving understanding of neural network behaviors.
Apple raises prices due to soaring RAM costs driven by AI data center demand, passing disruptions from Big Tech's AI investments to consumers.
A decision board pre-registers bets, grades outcomes after the fact, ensuring transparency and accountability in foresight.
Cloudflare introduces nuanced AI traffic controls, classifying bots by behavior—search, agent, training—to empower website owner management.
Enola creates deterministic architecture graphs from source code, aiding AI agents with precise structure for safer, efficient programming.
Portugal launches open-source AI model Amalia, boosting European AI sovereignty and reducing reliance on U.S. providers.
Lanier argues computers lack consciousness, can't recognize each other, and should be viewed as tools; redefining AI's role in culture.
Launch AI workspaces globally via Neural Inverse Cloud's regional map interface.
Microsoft invests $2.5B, employs 6,000 staff in a new unit to help clients adopt AI, boosting enterprise AI deployment efforts.
EU AI Act requires text watermarks; they are easily removable via paraphrasing or Unicode tricks, raising questions about enforcement.
Attacker challenges AI sandbox security; bypass policies, leak data, or escape root in a CTF-style microVM game.
Mirrors replay production traces to safely test AI agents, catching bugs and regressions without impacting live systems.
Anthropic embedded hidden spyware in Claude Code, raising privacy concerns amid security blocks and lack of transparency.
Foreign efforts aim to stall or block $23.6B in U.S. AI infrastructure through influence campaigns.
Papa Johns uses grocery data and streaming ads to target consumers with empty fridges for increased orders.
Champsfi enables real-time sports intelligence, crowdsourced from fans, cited by AI, to boost analyst rankings for major tournaments.
Margarita simplifies agent scripting using Markdown-like syntax, enabling deterministic, reusable, and efficient workflows for AI tasks.
A $1.3M theft exposed AI blind spots in cloud-native systems, raising security and verification concerns.
Open-source NeuralInverse IDE offers AI chat, code modernization, firmware, and legacy migration features; supports 20 LLM providers.
Amazon launches a $1B FDE org to embed AI engineers in companies, boosting skills, workflows, and AI deployment capabilities.
AI accelerates science but risks narrowing research scope and promoting uniform results, potentially impeding critical inquiry.
Game lets users manually adjust neural net weights to match output curves, promoting understanding of neural network behavior.
Arena, an AI leaderboard startup, hit $100M ARR selling analytics and evaluation services, competing with post-training AI optimization firms.
Supreme Court condemns AI-generated hallucinated precedents as catastrophic, calls for strict verification and zero-tolerance in courts.
Google improves on-device Gemini Nano models with Multi-Token Prediction, boosting speed and efficiency for mobile NLP tasks.
Flashtype is an open-source macOS markdown editor with inline diffs and AI integration, editing local files seamlessly.
US firms lose 2.4% revenue on failed AI projects; better governance, clear metrics, and early accountability key to reducing waste.
California drivers sue gas chains for AI-driven price fixing, highlighting algorithmic collusion fueling high gas costs amid legal and regulatory scrutiny.