AI & Machine Learning

From Earth to Orbit, France’s Ecosystem Delivers Agentic Computing at Scale With NVIDIA

Companies in France are deploying agentic computing across industries spanning space, financial services, the public sector and scientific research, demonstrating how the country’s ecosystem has moved from ambition to application. France’s progress sits on deliberate groundwork. The AI Action Summit in February 2025 and Choose France Summits in 2025 and 2026 anchored major compute commitments

A CNN on every frame caught 2 of 54 wildfire windows at a fixed alert budget. One 20-line baseline made it 18.

Wildfire lookout networks have far more camera feeds than anyone can watch. The useful question isn't "is there smoke in this frame?" It's "which of these dozens of tiles should a person open next?", asked without burying them in alerts until they mute the tool. I built PlumeRank around that question for the OpenCV AI Competition 2026. The most useful thing I learned wasn't in my own pipeline. It was what happened when I gave a baseline the same treatment. Live, no login: https://live.plumerank.edycu.dev · Code: https://github.com/edycutjong/plumerank Every system in the comparison gets the same allowance: 2 alerts per camera per 81-minute window, which is the length of one FIgLib sequence. Each one is calibrated to spend that allowance on a fire-free confusion set of 33 windows (low cloud, cloud shadow, overcast, cameras hunting exposure), and only then scored on fire footage. The allowance is enforced by a small, deliberately strict object: def spend(self, camera_id: str) -> bool: """Spend one alert on this camera. False when its allowance is already gone.""" if camera_id not in self._spent: raise KeyError(f"{camera_id!r} is not on the roster this budget was built for") if self._spent[camera_id] >= self._per_camera: return False self._spent[camera_id] += 1 return True An unknown camera raises instead of being auto-registered. A typo that silently created an extra camera would inflate the denominator that the whole comparison rests on. The threshold isn't chosen by hand. solve_threshold finds the lowest value whose realised alert rate on the confusion set stays within the budget. A system can't look fast by alerting more, because it isn't allowed to. My first baseline was the obvious one: run the same small CNN on every frame and alert when its probability crosses the solved threshold. At the fixed budget it detected 2 of 54 held-out fire windows. The reason is in the table footnote: the solved threshold landed within 1e-6 of 1.0. On a fire-free set full of cloud a

IDC: Nvidia's AI chip shipments in China fell 44% YoY to 678K in H1 2026, as its market share slid to 34% amid US export curbs and rising domestic competition (Reuters)

Reuters: IDC: Nvidia's AI chip shipments in China fell 44% YoY to 678K in H1 2026, as its market share slid to 34% amid US export curbs and rising domestic competition — Nvidia's (NVDA.O) share of China's market for data-centre AI chips fell to 34% in the first half of 2026, as US export controls restricted supplies …

LocalBiz AI: Turning Small Business Ideas into Ready-to-Use Content

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle What I Built I built LocalBiz AI, an AI-powered business assistant designed for freelancers, independent sellers, and small business owners who want to create professional business content without spending hours writing it themselves. Running a small business involves much more than delivering a product or service. Business owners also need to write product descriptions, prepare service listings, create promotional content, respond to customer enquiries, and keep their communication professional. For someone managing everything alone, these repetitive tasks consume valuable time. LocalBiz AI aims to simplify that process. Users provide a few basic details about their product or service, including its name, key features, target audience, and preferred tone. The AI then helps transform that information into useful business content, such as: Product descriptions for online stores. Service descriptions for freelancers. Promotional content for social media. Short marketing copy for advertisements. Professional responses to common customer enquiries. Alternative versions of content for different audiences. The focus is on small, practical tasks that business owners regularly encounter. Instead of paying for multiple writing tools or hiring someone for every small content requirement, a business owner can use one assistant to speed up routine writing. The potential customers are freelancers, local retailers, independent creators, home-based businesses, and small teams that need affordable content assistance. The idea behind LocalBiz AI is simple: make useful AI capabilities accessible to businesses that cannot justify expensive software subscriptions. How I Built It The project is designed around open-weight AI rather than relying entirely on a closed, commercial AI API. The proposed technology stack consists of: Frontend: Next.js and React. Programming language: TypeScript. UI styling:

Introducing AIGoat 2.0: Hands-On Security Labs for LLMs, RAG, MCP, and AI Agents

AI security is evolving rapidly. As applications move beyond standalone Large Language Models (LLMs) to Retrieval-Augmented Generation (RAG), connected tools, and autonomous AI agents, the attack surface is expanding. Prompt injection is only one part of the problem. Tool misuse, excessive agency, untrusted context, and unintended actions introduce new challenges for security engineers and developers. Understanding these risks requires more than reading about vulnerabilities. We need environments where we can reproduce attacks, investigate their impact, and experiment with defenses. That is the motivation behind AI Goat 2.0. We’re excited to introduce the latest evolution of AI Goat, an open-source, hands-on AI security playground designed to help practitioners explore how AI systems can be attacked and defended. AI Goat includes hands-on MCP security labs that help practitioners investigate risks associated with tool interactions, untrusted inputs, and the capabilities exposed to AI systems. AI agents can do more than generate text. They can select tools, interact with application functionality, maintain context, and perform multi-step tasks. This creates a different security challenge: an attacker may manipulate an agent's behavior rather than simply influence its final response. AI Goat's agentic security labs provide practical scenarios for exploring these risks and understanding their implications. Agentic Kill Chain challenge. The scenario explores how malicious instructions embedded in seemingly legitimate content can influence an AI agent and potentially lead to unintended actions. Instead of treating every vulnerability as an isolated issue, the challenge encourages practitioners to investigate how weaknesses can combine across an agent's context, tools, memory, and actions. This is particularly relevant when evaluating AI assistants connected to business applications, where an apparently harmless interaction can have consequences beyond the model's respons

How to Negotiate an AI Vendor Termination for Convenience Clause

Vectra AI, the network detection and response company, restructured its enterprise contracts in 2024 to replace flat early-termination fees with usage-based wind-down credits tied to actual compute and support hours consumed.

Meta claims new Muse Wonder AI models rival OpenAI, Anthropic

Meta co-founder Mark Zuckerberg said its newest models are a step toward “personal superintelligence.”

Experts launch Europe-wide framework to guide AI use in mental health

The first Europe-wide expert working group on Artificial Intelligence in mental health launches today to the ECNP congress in Munich, where the group will host an open discussion, setting out the potential, the problems, and possible directions for AI in psychiatry and neuroscience.

Preserve research data for tomorrow’s AI-enabled integrity checks

Is this idea from AI or humans? A detector claims it can tell

Inundated AI conferences impose submissions caps

Decentralized AI needs more evidence on energy use

In the AI era, coding means thinking like a systems architect

Leading Chinese mathematicians launch an alliance, including eight orgs and aiming for 1,000+ scientists, to focus on developing specialized AI models for math (Chao Kong/South China Morning Post)

Chao Kong / South China Morning Post: Leading Chinese mathematicians launch an alliance, including eight orgs and aiming for 1,000+ scientists, to focus on developing specialized AI models for math — Leading China's mathematicians joined forces in Beijing on Sunday to launch an alliance aimed at integrating artificial intelligence into mathematical research.

Google's Trends API is still invite-only. I tested what the alternatives return

TL;DR: Google announced an official Trends API in July 2025 "only to a very limited number of testers". Fifteen months later the page is still titled "Alpha" and still ends in an application form. pytrends, the library most tutorials use, was archived in April 2025, and its last maintainer left a note saying Google gives bots "slightly altered data". I tested that claim on 13 October 2026: five keywords, one hour, three ways to ask Google. Interest over time and interest by region came back the same from all three, give or take 3 points. Related queries are unstable for everyone, including the website. Related topics are the one thing Google withholds from anything it tags as a scraper, and it tags sessions openly, in its own JSON. What you can't get from any route below: absolute search volumes. Google Trends only ever gives relative interest on a 0 to 100 scale, API or not. Source State today Official Google Trends API Alpha. "As we can't open the API for everyone yet, we're prioritizing developers that know what they want to do", with an application form (API page). Announced 24 July 2025 (blog). An April 2026 Ask HN asking who got in has one reply, from someone who didn't pytrends (Python) Archived on GitHub 17 April 2025. Last release 4.9.2, April 2023. Still 908,810 PyPI downloads in the last 30 days (pypistats). 63 issues and PRs in the repo mention HTTP 429 trendspy (Python) The fork the pytrends maintainer pointed to. Last commit 25 December 2024, v0.1.6 google-trends-api (npm) Last push November 2022. 569,469 downloads in the last 30 days, 54 open issues trends.google.com Works. Compare a handful of terms, download a CSV by hand Scrapers (self-built or hosted) Work, with the caveats measured below The pytrends note is worth reading in full (issue #636, 8 February 2025). The maintainer's claim is that when Google detects a bot it returns "slightly altered data" compared with the website. That's a strong claim, and nobody I could find had checked it. So I di

Alibaba's Qwen team open sources QwenGyre to train AI agents that run for hours

QwenGyre cut training time by 44% on Alibaba's 2.4 trillion parameter Qwen3.8 model, completing 48 steps in 75.42 hours instead of 134.55, while lifting its NL2RepoBench coding pass rate to 58.54%.