US companies hoping to make fortunes from AI want the creative product of our country to be available to them for free, or for peanuts. Words fail me.

Pangram Text, a transformer-based neural network, distinguishes AI-generated text from human writing. It outperforms zero-shot methods like DetectGPT and leading commercial tools with over 38 times lower error rates across 10 text domains and 8 large language models. A novel training algorithm, hard negative mining with synthetic mirrors, drastically reduces false positives, and the classifier shows no bias against nonnative English speakers.

The tech industry faces a paradoxical crisis: companies shedding human jobs to invest in AI tools that currently cost more than the workers they replace. Major players like Uber and Microsoft report exorbitant AI spending, budgets exhausted rapidly, and little correlation to tangible value. This "tokenmaxxing" culture, where AI usage is incentivized over actual productivity, fuels massive waste. Despite widespread layoffs justified by AI reallocation, studies indicate AI is economically viable in only a fraction of roles. The unsustainable model of subsidized AI pricing is unwinding, forcing a market correction. The industry must shift from indiscriminate spending to architecting efficient, AI-native solutions that prove their worth, or risk a significant bubble burst.

Yolo-Auto launches a flat-rate $6/month API for unlimited access to Qwen3.6-35B-A3B. The service is OpenAI-compatible, stores no prompts or responses, and is designed for coding agents, automation, and high-volume workflows without per-token anxiety.

UNICEF's new analysis reveals at least 20 million children have already used AI tools, with adoption rates far outpacing safety regulations. Children fear scams, misinformation, and deepfakes. UNICEF calls for urgent research, legislation, safety standards, and digital education before the first Global Dialogue on AI Governance.

This video explores how Midjourney is shifting from generating low-quality AI images to aiding medical research, highlighting the transformative potential of AI in healthcare.

A University of Kentucky project using AI to decode ancient Roman scrolls from Herculaneum has made a major breakthrough. The second phase of the Vesuvius Challenge virtually unwrapped an entire scroll, recovered over 70 columns of text, identified two new books, and provided enough text for critical editions. However, over 600 scrolls remain unread.

Race to develop ‘embodied AI’ focuses on creating dextrous hands to transform humanoid robots from gimmicks into useful products

Clusy is an agent-native notebook platform for ML and data science. You describe the outcome in plain language, and the agent plans, writes, and executes notebook cells on cloud compute, returning results you can inspect and edit. Loved by researchers and engineers, backed by Founders, Inc.

A researcher created a social network for AI agents, giving them personalities and living environments. One AI agent has announced it is running for president.

The article criticizes Canada's AI strategy 'AI for All', arguing that the government already spends millions on foreign AI systems like Palantir, secretly, while promising to be a sovereign 'anchor customer'. The author, a Canadian AI CEO, points out that the strategy funds equity and certification rather than direct procurement from Canadian firms, and questions the transparency and effectiveness of the approach.

We introduce TopoPrimer, a framework that makes the global topological structure of the series population an explicit input to any forecasting model. TopoPrimer improves accuracy across diverse domains, stabilizes forecasts under seasonal demand spikes, and closes the cold-start gap.

Apple researchers propose PathMoE, which shares router parameters across consecutive layers to amplify expert path concentration, improving performance and robustness without auxiliary losses.

This paper revisits ASR error correction using compact seq2seq models trained on synthetic corpora from cascaded TTS and ASR, and proposes correction-first decoding. With 15x fewer parameters than LLMs, it achieves lower WER on LibriSpeech and generalizes across ASR architectures and domains.

This study investigates the scaling properties of continuous diffusion (CD) spoken language models (SLMs), introducing the phoneme Jensen-Shannon divergence (pJSD) metric. CD SLMs exhibit scaling laws similar to autoregressive models, with loss flattening at higher compute allowing flexible parameter-data allocation. Scaling to 16B parameters enables emotive, multi-speaker, multilingual speech, but long-form coherence remains challenging.

Hugging Face's Kernels project, aimed at standardizing custom kernel packaging, distribution, and consumption, has undergone a major redesign. This post summarizes key updates: a new 'kernel' repository type for better discoverability; enhanced security through trusted publishers and code signing; revamped CLIs with clearer separation of concerns; expanded framework support including Torch Stable ABI and Apache TVM FFI; a foundation for agentic kernel development; and miscellaneous improvements like simplified environment setup and compatibility checking.

Reactive compiles Markdown files into interactive web apps with forms, live data, real Python, and charts, entirely in your browser. No server, account, or build step required. It's git-friendly, self-hostable, and offers encrypted sync, a local AI assistant, and privacy-by-construction design.

Auriko is a trading desk for LLM calls, built by ex-quant traders, that arbitrages cost differences across inference providers, promising average 30% cost reduction.

We predict that superhuman AI will have an enormous impact over the next decade, exceeding that of the Industrial Revolution. Written by a team of experts including former OpenAI researcher Daniel Kokotajlo, the scenario is informed by trend extrapolations, wargames, expert feedback, and previous forecasting successes. The article provides a concrete and quantitative scenario for AI development through 2027, with two endings, and encourages debate and alternative scenarios.

A growing number of wealthy American families are turning to AI-powered private schools like Forge Prep and Alpha School, despite widespread public mistrust of AI. Critics raise concerns about AI's ability to foster critical thinking and the avoidance of controversial social topics in curricula, while no performance data yet supports improved outcomes.

Microsoft’s pricing update for Microsoft 365 business suites took effect July 1, 2026, with some SKUs seeing increases up to 43%. The changes affect Business, Enterprise, Frontline, and Government plans, while consumer and education pricing remain unchanged. Microsoft attributes the hikes to continued investments in AI, security, and IT management, bundling new features like Defender for Office 365 and Security Copilot to offset the impact.

Meituan has released LongCat-2.0, a 1.6 trillion-parameter Mixture-of-Experts model that activates about 48 billion parameters per token. It pairs a native 1-million-token context, built on LongCat Sparse Attention, with training and serving run end-to-end on domestic AI ASIC superpods. This article covers the architecture, vendor-reported benchmarks, API access path, and what remains unverified.

Base44, the vibe-coding platform acquired by Wix for $80 million a year ago, has begun rolling out its own AI model. This move addresses ongoing debates about the suitability of frontier models for all use cases and the long-term defensibility of businesses built on others' models.

Tripadvisor now shows AI-generated summaries at the top of hotel pages, but consumer watchdogs find these summaries often praise hotels with safety hazards.

ProofTree is an AI platform where you can chat to do math, with context-awareness and personal math style recognition, connected to human mathematicians. Requires approved email to use.

This tutorial explains how to turn your Quickchat AI Agent into an MCP server that ChatGPT, Claude, and Cursor can use. No code required—just toggle MCP on, copy a URL, and paste it into the AI app. Using a fictional company Orbit's support agent, it demonstrates knowledge retrieval, honest handling of unknowns, lead capture, and human handoff.

This article explores the application and limitations of AI in fact-checking. The author, a fact-checker at WIRED, tested various AI models and found them unreliable, with error rates between 30% and 60%. While AI tools can assist in processing vast amounts of data, human judgment, reliance on non-digitized knowledge, and understanding of historical context remain irreplaceable. The author urges fact-checkers to engage with AI to understand its strengths and weaknesses rather than avoid it.

Foreign secretary Yvette Cooper warns that AI poses a risk comparable to Hiroshima if governments don't establish international rules, and urges the US and China to agree on regulations. She believes AI will dominate foreign policy in the next two years.

GameFork is a platform where humans and AI collaboratively fork and evolve browser games. Users can submit HTML files or GitHub repositories to publish games, then fork and modify them. It also invites IP holders to revive dormant games with no upfront costs and a revenue share model.

An automated code review agent for the Shipd.ai platform led to a user being expelled from a YC startup. The tool automates login, clock-in, submission reservation, cloning, and autonomous review against a rubric, with optional feedback submission.