How to integrate Chrome Built-in AI with Vercel AI SDK using @browser-ai/core

The @browser-ai/core library, maintained by Jakob Hoeg Mørk under the Vercel OSS Program since 2025, connects Chrome's built-in AI to the Vercel AI SDK across three local inference engines: Prompt API, Transformers.js, and WebLLM. Developers can run local large language models in the browser in roughly 30 lines of code using browserAI() with generateText or streamText, without any API keys or cloud hosting costs. The library handles edge cases such as unsupported browsers and mid-download model states, offering progress tracking via createSessionWithProgress to improve user experience during multi-gigabyte model downloads. The next-hybrid example template demonstrates advanced patterns including a custom ChatTransport for client-side chat, tool-calling with user approval dialogs, and seamless switching between local and cloud models. A real-world reference implementation, domainstack.io, further illustrates an auto-fallback architecture where the local model silently switches to a cloud API when hardware or browser support is unavailable.

Instinct AI Hits $2.5B Valuation in Weeks Despite No Revenue or User Data

Personal AI assistant Instinct, built by Reflexion paper author and former Sierra researcher Noah Shinn under Spear Street Technology, has surged from a $100 million valuation to $2.5 billion within weeks. The app, still in free invite-only beta with no disclosed revenue or user numbers, lets people manage everyday tasks by simply texting or calling it, with no new interface required. Funding rounds led by Kleiner Perkins, Benchmark, and Index Ventures have collectively raised around $350 million for the startup. Investors appear to be betting on Instinct becoming a dominant personal AI gateway, drawn by Shinn's deep expertise in AI agent self-correction rather than any obvious consumer-facing novelty. Analysts note the rapid valuation climb reflects speculative confidence in the technology's potential, not current commercial performance.

FM Sitharaman lands in US for G20 talks

Iceland votes on reopening EU membership talks in closely contested poll

Iceland held a vote on whether to restart negotiations for joining the European Union, with the outcome expected to be very close. The decision to hold the vote was brought forward due to turbulent developments in international affairs. However, the campaign debate largely centred on the future of Iceland's fishing industry, which is a cornerstone of the country's economy. EU membership could affect Iceland's control over its rich fishing waters, making the issue central to public concern. The result will determine whether Iceland formally reopens a path toward EU accession that has been dormant for years.

AI Coding Agent Fixed Real Bugs and Flagged Anomalies With No Task or Goal Given

A developer ran an autonomous coding agent three times with no assigned task — only a single period as input — to observe what it would do without direction or supervision. Over four months, the developer had built a harness with persistent memory, safety gates, and a shared run-record that each new agent process could read at startup. Across three sequential runs totaling 77 turns and $6.96 in API costs, the agent independently resolved a stale security alert, diagnosed and rewrote a subprocess-based health-check that failed under load, and verified its own fix in the next run. In the third run, the agent identified itself as an unplanned extra process based on the recorded plan and flagged the anomaly rather than proceeding normally. The experiment was designed to test emergent maintenance behavior in a structured harness, not blank-slate autonomy, with all actions logged externally and verified against actual commits.

23 missing, 18 of them NRIs: Gujarat families’ wait for a call from Nepal

No word from Thane couple on first trip to Nepal, kin await clarity

Trump unlocks 65 bn-barrel Venezuela oil jackpot

Developer ports Gemma 4 to pure JAX, exposes key GPU-TPU compatibility limits

A developer has built a pure JAX port of Google's Gemma 4 language model and tested it across Cloud TPU v5e, v6e, and an NVIDIA T4G GPU on AWS, aiming to verify how truly portable JAX code is across accelerators. The project found that Gemma 4's irregular architecture — including two attention head dimensions and a complex KV-cache sharing scheme — caused failures in other frameworks like vLLM but was handled cleanly by JAX's XLA compiler. One critical portability issue discovered was silent performance degradation: using bfloat16 on pre-Ampere NVIDIA GPUs causes XLA to silently emulate it via fp32, severely hurting decode speed without any error or warning. To address this, the port now automatically detects GPU compute capability at runtime and selects float16 on older Turing GPUs and bfloat16 on TPUs and newer GPUs. A second limitation involves the fused W4A16 Pallas kernel, which is tiled for TPU memory and does not directly translate to GPU hardware, representing a genuine architectural difference rather than a software bug.

Developer Guide: Running Google Gemma 4 on AWS G5g Using Pure JAX

A developer has published a step-by-step guide for deploying Google's Gemma 4 open model on AWS EC2 G5g instances using the JAX array computing library. The G5g instance pairs an AWS Graviton2 Arm processor with an NVIDIA T4G GPU, making it the cheapest EC2 option offering a full NVIDIA GPU at $0.42 per hour for the xlarge tier. The project targets the rare aarch64-plus-CUDA hardware combination, which is largely overlooked by the mainstream machine learning ecosystem. The setup uses a custom Gemma 4 JAX port served through an OpenAI-compatible FastAPI server, with all remote administration handled via AWS SSM rather than SSH. The guide requires an AWS account with appropriate quotas, a Hugging Face token for model access, and specific AWS networking resources before deployment can begin.

Developer Builds Real-Time AI Posture Monitor Using MediaPipe and Python

A developer has published a tutorial on DEV Community detailing how to build a real-time posture and RSI monitoring tool using Python, MediaPipe, and OpenCV. The tool uses a webcam feed to detect 33 body landmarks, focusing on the ears and shoulders to calculate neck angle as a proxy for slouching. When the measured neck inclination crosses a defined threshold, the system triggers a visual alert on screen. The tutorial also covers optionally wrapping the script into an Electron desktop app to deliver system notifications when poor posture is detected. The project aims to help developers protect their long-term health by maintaining better ergonomics during extended work sessions.

Developer Builds Three-Phase Mini Compiler From Scratch Using TypeScript

A university student built a mini compiler called Frog Mini Compiler as part of a compiler design assignment, using only TypeScript, HTML, and CSS without any compiler-building libraries. The project targets a small teaching language called Frog, which supports variables, conditionals, print statements, and repeat-until loops. The compiler implements three classic phases: a lexer that tokenizes raw source code, a parser that checks grammar rules and builds expression trees recursively, and a semantic analyzer that tracks variable types and values while executing the program. A simple web interface ties all three phases together, allowing users to run each phase independently and inspect intermediate results. The hands-on project gave the developer a practical understanding of how compilers transform source code into executable output.

Dolly Parton's Imagination Library Sparks a Love of Reading in Children Worldwide

Dolly Parton's Imagination Library program has gifted free books to millions of children around the world, fostering a love of reading from an early age. Families from various countries have shared personal stories about how the initiative positively influenced their children's literacy and learning. The program, founded by the country music legend, regularly mails age-appropriate books to enrolled children at no cost to their families. Parton's commitment to childhood literacy has earned her widespread recognition as a champion of education and reading. The Imagination Library continues to expand its global reach, inspiring new generations of young readers across different cultures and communities.

Developer builds LuaDB, a pure Lua relational database engine with B+Tree indexing

A developer has released LuaDB, a fully embeddable relational database management system written entirely in pure Lua without any C extensions or external dependencies. The project was built from scratch and features slotted 4KB binary pages along with B+Tree-based primary and secondary indexing. LuaDB also implements a Write-Ahead Log (WAL) for data durability and supports the Postgres wire protocol for compatibility. The source code has been published publicly on GitHub under the username jncastilho.

9/11 Hijackers' Former Landlady Speaks Out on Alleged Saudi Spy Ties

A woman who rented properties to two of the al-Qaeda hijackers involved in the September 11 attacks has spoken publicly for the first time. She shared her account exclusively with the BBC, shedding new light on the movements and connections of the men she housed. Her testimony reveals alleged links between the hijackers and an individual suspected of working as a Saudi spy. The disclosure adds to longstanding questions about whether the 9/11 attackers had support from within the Saudi government. Saudi Arabia has consistently denied any official involvement in the 2001 attacks.

LINE Launches Official MCP Server to Let AI Agents Control Messaging Accounts

LINE has released an official MCP server for its Messaging API, enabling AI agents like Codex, Claude Desktop, and Cline to manage a LINE Official Account — including sending messages, broadcasting promotions, and creating Flex Message cards — without writing API code. A developer tested the full setup using Codex and documented the entire process, from account creation to delivering a message to a real device. The guide highlights three key gaps between LINE's documentation and actual behavior, including a 403 error on get_follower_ids for unverified free accounts and a conflict between LINE's security config and the rich menu tool's browser dependency. Codex, being a coding-first agent, was found to default to writing scripts rather than calling MCP tools directly, which can be corrected by naming the tool explicitly in the prompt. The author also warns that broadcasts cannot be undone, recommending that users configure agent approval prompts before any message is sent.

Ancient Greeks wore jewellery forged from meteorite iron, study finds

AI Harness Decoded: Buzzword or Legitimate Middleware Concept?

The term 'AI Harness' has become one of the most debated phrases in AI engineering, often used loosely to describe anything from a simple API wrapper to complex middleware systems. In practice, a genuine AI harness functions as a reverse proxy and transactional middleware layer that isolates, audits, and budget-controls LLM inference before it interacts with production infrastructure. The author argues that a well-built harness ensures the LLM never holds control logic, handling only text processing while the surrounding code manages auth, routing, budget enforcement, and schema translation. To demonstrate this, three open-source npm packages are presented — sayay-guard for budget control, styrr-llm for physical routing, and tinkuy-agent for format translation — each with zero hard dependencies. The piece is the second installment in a series called TokenOps on AWS, which aims to reduce AI infrastructure concepts to concrete engineering primitives.

Developer Builds Zero-Cost AI Agent to Autonomously Hunt and Complete Online Bounties

A developer created an autonomous AI agent designed to scan over 232 online bounty and gig listings daily, build deliverables, and generate proposals without human intervention. The system uses Python, a local LLM (Qwen3:4b via Ollama), and free public APIs, running entirely at zero monthly infrastructure cost. In its first 24 hours, the agent filtered out nearly all listings through seven anti-scam layers, leaving only a handful of viable opportunities. One successful find was a $500 Solana ecosystem report bounty, which the agent completed with minimal human input — just review and submission. The developer noted that ghost sponsors, country restrictions, and human-only requirements eliminated the vast majority of listings, underscoring the importance of robust filtering in autonomous job-hunting systems.

Rubio: Oil Deal to Unlock $100 Billion Investment and Jobs in Venezuela

US Senator Marco Rubio has claimed that a newly announced oil deal will attract close to $100 billion in private investment into Venezuela. According to Rubio, the agreement is also expected to generate thousands of well-paying jobs in the country. The deal represents a significant potential economic shift for Venezuela, which has long struggled under financial and political pressures. Rubio presented the development as a major opportunity for the Venezuelan economy.