AI Boom Mirrors Railway and Dot-Com Eras: Real Tech, Real Bubble Risk

Analysts drawing on historical parallels argue that 'real technology' and 'speculative bubble' are not mutually exclusive — railways and the internet were both transformative and financially ruinous for many early investors. The four largest US hyperscalers are collectively spending hundreds of billions of dollars annually on AI data centres, with capital expenditure revised upward nearly every quarter, mirroring the massive physical infrastructure bets of the railway age. A key concern is that Nvidia, the dominant AI chip supplier, has taken stakes in and provided financing to some of its own largest customers, creating a reflexive loop where chip purchases and valuations reinforce each other rather than reflecting independent outside demand. This dynamic echoes the 1720 South Sea Bubble, in which the company lent buyers money to purchase its own shares, making rising prices appear as validation of the scheme. The more useful question, historians and analysts suggest, is not whether AI is a bubble, but which parts represent durable infrastructure build-out and which parts are speculative financing structures dressed in technological clothing.

Jev Targets AI Decision-Making as a Distinct Layer from Text Generation

A new AI tool called Jev is designed specifically for decision-making workflows rather than generating text responses. Unlike conventional large language models, Jev accepts application state and typed questions, then returns structured outputs such as routing decisions, probabilities, and confidence scores. The approach addresses use cases like support ticket routing, sales lead prioritization, agent action retries, and content review pipelines. The core architectural idea is to treat decision-making as a separate concern from language generation, allowing applications to consume clean, typed outputs directly. As AI systems shift from chat interfaces toward continuous action-taking workflows, this decision-oriented model pattern is being watched as a potentially better fit for such applications.

Sports platform unlocks worldwide live scores by fully using its paid API key

A niche sports database site, nichedb.dev, has added a second live score source after discovering its existing paid TheSportsDB subscription offered far greater data access than its code was using. The subscriber API key unlocks 1,544 leagues in a single request, full team rosters with 63 fields, and a global live scores endpoint — capabilities the site's code had never utilized despite the key already being active in production. The new live score feed returned 80 games on its first production run, covering leagues such as Slovak Extraliga, Finnish Liiga, and African Cup of Nations qualifiers. The existing ESPN-based source is retained alongside the new one, as it provides richer match detail like play-by-play and venue data, though it requires a costly residential proxy due to cloud access restrictions. The broader lesson highlighted is that paid API keys are often deployed without revisiting the workarounds written for free-tier limitations.

Nvidia CEO Jensen Huang argues AI's energy costs are worth long-term climate gains

Nvidia CEO Jensen Huang claimed in an interview on The Ezra Klein Show that AI can ultimately help address climate change. However, he acknowledged the technology would first cause significant environmental harm due to energy-intensive data centers. His stance mirrors an accelerationist view shared by several tech executives and political figures, including President Donald Trump. Critics argue this position downplays the immediate environmental damage caused by AI infrastructure running on fossil fuels. The Verge noted that Huang's comments reflect a level of privilege that overlooks the real-world impact on communities affected by climate change today.

Parrot

Discussion | Link

Developer Releases FastAPI Microservice to Clean CSV Data Using Pandas and Docker

Developer Marcela Zapata Vanegas has published an open-source Python microservice that ingests messy CSV files, cleans the data using Pandas, and returns structured JSON responses. The project combines FastAPI for asynchronous API handling, Pandas for data manipulation such as resolving missing values and normalizing headers, and Docker for containerized deployment. The containerized service runs on port 8000 and can be deployed on major cloud platforms including AWS and DigitalOcean. FastAPI's built-in Swagger UI allows users to test the cleaning endpoint directly in a browser without needing external tools like Postman. The source code is publicly available on GitHub under the repository proyecto-fastapi-csv.

Developer Builds Tiny Prompt Journal to Block Redundant AI Model Requests

A developer working on a text-classification lab in Halifax repeatedly sent identical prompts to an AI model after making only mental notes rather than actual changes to the request payload. The experience revealed a common pitfall: mistaking a revised thought for a revised input, resulting in the same wrong output each time. To address this, the developer built a lightweight Python tool that hashes and journals each prompt request, blocking any send where the diff between the current and previous request is empty. The tool relies solely on Python 3.9+ standard library modules — hashlib and json — requiring no third-party packages. It signals blocked attempts via stdout and a shell-readable exit code, making failures more visible than the sender's optimism.

Election Commission verifies VIP voter entries for Delhi's final electoral rolls

The Election Commission of India has validated the electoral entries of prominent voters in New Delhi for inclusion in the final voter rolls. These individuals received Special Intensive Revision notices requiring verification of their personal details. The Commission stated it followed standard procedures in processing these verifications. Responses to the notices were accepted through both online and offline channels. The move highlights the routine but closely watched process of maintaining accurate electoral records in the capital.

NASA Awards $300,000 Top Prize in Mars Food Systems Design Challenge

NASA announced the winners of its Deep Space Food Challenge: Mars to Table on Thursday, recognizing innovative designs for integrated space food systems. The top prize of $300,000 went to Chinyere Ukeje of Philadelphia, Pennsylvania, for a concept called the Adaptive Nourishment Infrastructure (ANI). The competition tasked participants with developing solutions capable of providing safe and nutritious meals for deep space missions. The challenge was part of NASA's broader effort to address food sustainability for future crewed missions to Mars.

Hippocampus-Inspired Memory Architecture Aims to Fix Context Limits in AI Coding Agents

Large Language Models used in software development face a fundamental constraint known as the context window limit, which causes them to lose track of earlier information during complex, multi-step coding tasks. This leads to failures such as forgotten method signatures, drifting objectives, and prohibitively high computational costs when entire histories are re-fed at each reasoning step. Engineers are now drawing on neuroscience to address this, developing so-called Hippocampus architectures that separate an agent's working memory from a persistent long-term memory store built on vector databases and knowledge graphs. Modelled on the brain's hippocampal role in consolidating short-term experience into long-term recall, these systems allow coding agents to index and retrieve relevant context without overwhelming the active inference window. The approach is presented as a more scalable and economically viable path toward autonomous agents capable of handling large, production-grade codebases.

Why AI Model Outputs Need a Versioned Schema Before Touching Your Database

A software developer writing for DEV Community argues that free or low-cost AI model outputs should be treated as untrusted upstream data, similar to a flaky webhook, rather than reliable function calls. The author describes a recurring failure where a summarizer app returned HTTP 500 errors because raw model completions containing markdown fences were passed directly to a JSON parser without validation. The proposed fix is a versioned Pydantic schema pinned between the model's response and the database write path, so malformed or drifted outputs are rejected with a 422 error before any row is touched. The article includes a Python code example using Pydantic's BaseModel to enforce strict field constraints on summary text, schema version, and confidence score. The author also recommends logging rejected completions as failed attempts, so on-call engineers retain a traceable record rather than facing a bare 500 error with no context.

Most Enterprises Scale AI Agents Before Defining Measurable Business Outcomes

A growing number of enterprises are deploying AI agents at speed without first redesigning the workflows those agents are meant to improve, according to recent industry analysis. IBM research found that 64% of surveyed CEOs cited fear of falling behind as the primary driver of AI investment, even before value is established, while only 25% of AI initiatives delivered expected returns. McKinsey's 2025 survey similarly found that while 62% of organizations were experimenting with agents, only 23% had scaled an agentic system anywhere in the enterprise. Experts argue that autonomy should be granted through a defined 'delegation envelope' covering one actor, one goal, an allowed action set, an evidence contract, and a recovery boundary. The core recommendation is that enterprises should first identify measurable outcomes — such as reducing account-opening time or detecting fraud — and then determine what authority each part of a workflow has actually earned, rather than treating agent deployment as a goal in itself.

Print’s ad resilience gets a festive season boost

‘Let Gaza live’: Netanyahu faces protests in New York as demonstrators block roads ahead of UN speech

Macklemore announces 'Free Palestine' tour with proceeds for Palestinian aid groups

American rapper Macklemore has announced a tour titled 'Free Palestine', with the first show scheduled in Dublin next month. The rapper stated that all proceeds from the tour will be donated to organisations supporting Palestinians. The announcement follows a reported fallout between Macklemore and British singer Ed Sheeran. Macklemore has been a vocal advocate for Palestinian rights, and the tour appears to be an extension of that activism.

Emilio Gay century powers Durham to commanding total at Old Trafford

Durham batter Emilio Gay scored a century to set a strong foundation for his side in their County Championship match against Lancashire at Old Trafford. The innings helped Durham, the Division Two champions, post a substantial total in the contest. Lancashire's fielding let them down during the innings, with dropped catches proving costly for the home side. Gay's knock underlined Durham's quality as they look to make an impression following their Division Two title win.

DocumentDB 0.116 Adds Distinct Scan for $group Queries, Cutting Index Reads Sharply

DocumentDB 0.116, released on August 20, 2026, introduces a distinct scan optimization for MongoDB-style $group aggregation queries running on its PostgreSQL extension. Instead of scanning all index entries and discarding duplicates, the new path jumps directly between distinct key values, reducing index reads from 50,000 entries to just 100 in a test dataset. The feature, called enableGroupByDistinctScan, is disabled by default and must be explicitly enabled via a system setting. DocumentDB is an open-source PostgreSQL extension that implements the MongoDB API, with Microsoft as its primary contributor through feedback from Azure DocumentDB enterprise users. The optimization builds on the same loose-index-scan principle previously applied to MongoDB's DISTINCT_SCAN for first- and last-per-group queries.

Goa CEO assures action after 97 voters found missing from electoral rolls

Goa's Chief Electoral Officer has addressed concerns over 97 voters who were left out of the state's electoral rolls. The individuals were flagged during the Special Intensive Revision process due to data discrepancies. Subsequent door-to-door visits by officials helped re-register many of the eligible voters. The episode has cast doubt on the reliability of the Election Commission's ECINet software. Opposition parties have seized on the issue to criticise electoral registration procedures and the centralisation of voter data management.

Meta launches AI game-building tools for Horizon, with reach via Facebook and Instagram

Meta has announced two new AI-powered development tools aimed at helping users create games for its Horizon social platform. Horizon Create is a mobile app, while Horizon Studio is a browser-based tool offering more detailed controls. Both apps are entering early access, with users able to join a waitlist to get started. To address Horizon's struggles as a platform, Meta plans to give games built with these tools visibility on Facebook and Instagram, where they will also be playable.

DocumentDB 0.113 adds covering index support for $group aggregation pipeline

DocumentDB version 0.113, released on June 22, 2026, introduces an index-only access path for MongoDB-style $group aggregation queries within its PostgreSQL execution engine. DocumentDB is a fully open-source PostgreSQL extension that implements the MongoDB API, offering an open alternative for MongoDB applications. The new feature allows aggregation queries using $group and $sum to read directly from index entries rather than scanning full documents, improving query efficiency. In benchmark tests comparing versions 0.112 and 0.113 on a 10,000-row dataset, the updated version achieved zero documents examined while scanning only index keys. Microsoft is the primary contributor to the project, refining the extension based on enterprise customer feedback gathered through Azure DocumentDB.