pumaDB is a lightweight hosted memory layer for AI agents, offering shared memory via MCP or server-side API to store project context, research notes, transcripts, and more, without managing databases or vector stores.

DeepSWE v1.1 retains the same long-horizon engineering tasks as v1, but updates agent execution and scoring by grading committed code in an isolated environment for reproducible, auditable results. It adds Claude Fable 5 and Kimi K2.7 Code, with aggregate pass rates remaining close to v1.

The Department of Justice intervened in a Clean Air Act lawsuit against xAI, arguing that Elon Musk's Grok AI is vital for national security and should be exempt from pollution regulations. Residents near xAI's Memphis data center suffer from noise and air pollution from unpermitted gas turbines.

On June 12, 2026, the US government forced Anthropic's Claude Fable 5 model offline. This article analyzes four scenarios: sincere mistake, genuine capability danger, foreign access concerns, and politics. The author favors politics but provides probabilities and forecasts for each scenario.

The LF AI & Data Foundation has formed a working group to develop DocLang, an AI-native document format designed to replace PDFs and other formats for better machine consumption. Backed by IBM, NVIDIA, Red Hat, ABBYY, HumanSignal, and Forgis, DocLang maps document elements to LLM tokens 1-to-1, reducing costs and improving accuracy.

Meta announced the global availability of its AI agent (Meta Business Agent) on WhatsApp Business, capable of answering questions, recommending products, booking appointments, and more, after nearly two years of testing in India and Mexico.

The article discusses an invisible problem in AI interactions: why some land and others sour. Factors like drift, tone mismatch, and weird behavior are hard to address because they are locked in an inaccessible layer. The author proposes observation as the foundational layer under taste and has built it for AI runtime.

Modal is all-in on speculative decoding, arguing it's the single most important inference optimization, delivering 2-3x speedups. They released state-of-the-art DFlash speculators for Qwen models, achieving 5-20% extra speedups, and explain the theory, simulation, and math behind the acceleration.

Datasette Apps is a new plugin that lets users run self-contained HTML+JavaScript applications inside a tightly sandboxed iframe within their Datasette instance. These apps can perform read-only SQL queries and, with stored queries, write operations. The plugin leverages iframe sandbox attributes and Content Security Policy for security, uses postMessage and MessageChannel for locked-down APIs, and supports AI-assisted app generation via copyable prompts. The article discusses a security vulnerability fix involving CSP allow-listing, visible logging, and the broader vision for Datasette's evolution into a richer tool ecosystem.

AA-Briefcase is a new benchmark from Artificial Analysis that tests AI models on realistic multi-week knowledge work projects. It combines rubric and pairwise grading to evaluate task success, analytical quality, and presentation quality. Claude Fable 5 leads but is expensive; open-weight model GLM-5.2 offers strong price/performance.

Amazon SageMaker AI now emits over 100 detailed inference metrics covering GPU health, token-level latency, KV cache pressure, traffic distribution across Availability Zones, and more. These metrics are displayed in a built-in SageMaker Insights dashboard in CloudWatch, which supports PromQL queries. This post explains how to enable detailed observability, navigate the dashboard, and connect metrics to external tools.

Access 337+ MCP tools covering weather, crypto, finance, web search, DNS, colors, AI chat, text analysis, calculators, and more. Pay-as-you-go at $5 for 500 calls.

A new Carnegie Endowment paper argues that AI infrastructure location will shape global power, with 'time to power' being the most critical factor for attracting data center investments. Democracies must reform domestic processes and forge international coalitions to keep AI development aligned with liberal values.

Unsloth is a tool that lets users run and train AI models 100% offline on local Mac and Windows devices. It supports GGUF and Safetensors models with tool-calling, web search, and an OpenAI-compatible API. Users can train models without coding, compare models side by side, import documents to create datasets, and export models. Unsloth offers a free open-source version as well as paid Pro and Enterprise tiers.

This article examines the social and psychological impact of AirPods and earphones, noting a 28% decline in spoken words among Americans from 2005 to 2019. It explores how earphones reduce spontaneous interactions, increase loneliness, and alter perceptions of audio content. The author argues for reducing earbud use to allow for self-reflection and genuine human connection.

This article argues that contrary to fears that AI reduces cognitive effort, writing with AI actually requires students to engage in deeper thinking, as they must evaluate, edit, and integrate AI-generated content.

European AI startup Legora, focusing on legal services, recently raised $550M Series D at a $5.6B valuation. The company boosted brand awareness by featuring actor Jude Law in a viral ad campaign. Legora now exceeds $100M ARR, serving 1,000 law firms across 50 markets. CEO Max Junestrand credits ChatGPT's API as a pivotal moment and says they have only built 1% of the software they plan to create.

Palmier Pro is a video editor integrating AI generation directly into the timeline, supporting models like Kling, Seedance, Veo, and Grok. It offers multi-track editing, MCP agent connectivity, and export to Premiere and DaVinci.

Sycloop is an AI-powered marketplace where people exchange items, skills, services, and digital assets without relying solely on cash. The AI engine finds multi-party trades that users wouldn't find on their own. Optional verification, free listings, and paid extras for higher confidence.

Amazon Prime Day 2026 is offering many smart home deals. A ZDNET expert picks the best security camera bargains worth your time and money.

xAI launches Grok for Word, an AI-powered Microsoft 365 add-in that allows drafting, restructuring, and tightening text directly inside Word via a conversational sidebar. It features structured note formatting, inline edits with before-and-after comparisons, and undoable revisions. Free from Microsoft Marketplace.

This article describes a comprehensive memory architecture for conversational AI assistants, featuring short-term memory (STM) and long-term memory (LTM), along with an internal AEIL language for compression and security. The core principle is 'Lead vs Cotton': storing dense information internally while delivering a smooth experience to the user. The system includes an invisible context bypass, per-message timestamps, and an importance hierarchy to enhance efficiency, personalization, and safety.

This skills file, authored by Gibraltar barrister Philip Vasquez, is designed to help AI systems navigate Gibraltar's fragmented public information landscape. It covers the legal framework, official data sources, and research methodology, including the hierarchy of laws, court structure, civil procedure, and key data sources such as judgments, legislation, Hansard, FSC regulated firms, and the Gambling Division. The document emphasizes that transparency is fundamental to democratic governance and rule of law, and provides detailed guidance for accessing public records.

Oalabs researchers discovered a hacker abusing Anthropic's Claude AI model for malicious activities including exploit development, cryptocurrency theft, and access brokering. However, the hacker's operational security failure led to a personal resume being included in the prompts, thus exposing their identity.

OSymandias is a Python library and CLI that turns your project into a full multi-agent runtime. It manages PostgreSQL, Redis, RabbitMQ, and Qdrant via Docker, and provides a dashboard, API, tool decorators, agent support, context management, and more. Supports multiple LLM providers.

As companies ration AI access due to soaring bills, Axamy argues the right move is an unlimited subscription model that forces internal efficiency improvements, not usage-based pricing that damages trust.

Perplexity launched Brain, a self-improving memory system for its Computer agent. Unlike traditional memory that remembers user preferences, Brain remembers the agent's work—what worked, what failed, and what corrections were made. It builds a traceable context graph, reviews it overnight, and learns to improve performance, with early results showing +25% correctness, +16% recall, and -13% cost.

Amazon Prime Day runs June 23-26, with early deals on Garmin smartwatches. Learn how to shop wisely and find the best discounts.

In a consequential grid infrastructure decision, the Federal Energy Regulatory Commission (FERC) today issued a major milestone on large-load interconnection impacting how those building AI factories, semiconductor fabrication support systems and advanced manufacturing facilities can connect to the grid. The new framework streamlines the interconnection queue, allowing large customers to fund upgrades, bring new generation, and offer flexible load, with study periods as short as 60 days. Data shows that increasing consumption can lower retail prices. This policy promotes growth, affordability, and reliability.

Data scientists sit at the intersection of analytics, machine learning and AI, turning raw data into predictive models, experiments and recommendations that guide business decisions. This article explores the evolution of the role, core skills needed, challenges faced, and how modern platforms accelerate the path from exploration to deployment.