[AINews] Open Models, Model Labs vs Agent Labs, and What's Untrainable — Sarah Guo
a quiet day lets us reflect on a great essay
Latest AI/ML News
30948 matching items
Why AI hasn’t replaced software engineers, and won’t
Coding agents as normal technology
AI Weekly Issue #502: Your AI can now spend your money — Visa wired it into ChatGPT
Visa just wired ChatGPT to shop and pay on your behalf — an AI agent can now buy at any Visa merchant without you clicking "buy." It capped a week where the labs pushed autonomy and capital to new highs: Anthropic put Claude Fable 5, its most powerful public model, into everyone's hands; Jeff Bezos came out of stealth with Prometheus, a $41B startup building an "artificial general engineer." A self-replicating worm hit 73 of Microsoft's own GitHub repositories through AI coding tools. Anthropic broke with the White House over preempting state AI laws; a German court ruled Google is liable for what its AI Overviews say. The agents got more capable this week — and a lot more autonomous.
Vector Space Hackathon 2026
Wow. So many cool and creative submissions for this year’s hackathon; we really had a tough time picking only 3 winners! The submissions ranged from early mental health detection to crowd-reaction simulators, tactical football search, and infrastructure stress-testing. We’re excited to share the results with you. The Hackathon Qdrant’s 2026 “Think Outside the Bot” hackathon pushed the creative boundaries of vector search. Participants from around the world were challenged to create innovative uses of Qdrant, without the use of RAG or simple chatbots. Submissions were judged on the criteria of Innovation, Creativity, and Technical Depth. The hackathon ran for 5 weeks with winners announced at Vector Space Day 2026 with a total of $10k in prizes. Keep reading to learn about the winning submissions.
From virtual experiments to biomedical insight with synthetic data
Nature Machine Intelligence, Published online: 11 June 2026; doi:10.1038/s42256-026-01244-6 Synthetic datasets are becoming crucial for the development of biomedical machine learning models. Victoriano et al. discuss the persistent simulation-to-reality gap that limits how well synthetic performance predicts real-world performance.
Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language
Nature Machine Intelligence, Published online: 11 June 2026; doi:10.1038/s42256-026-01250-8 Xiong et al. introduce ConfSeq, a molecular conformation description language that enables language models to perform three-dimensional molecular modelling tasks, including conformer prediction, three-dimensional molecular generation and representation, with strong performance.
BBVA puts AI at the core of banking with OpenAI
Learn how BBVA scaled ChatGPT Enterprise to 100,000 employees and partnered with OpenAI to accelerate AI-powered banking transformation worldwide.
Google DeepMind releases DiffusionGemma, a model that runs local AI 4x faster
Diffusion AI is most common in image generation, but it can make text outputs much faster.
Cristian Herrera reported where fear tried to silence others. Then he was killed
Herrera, despite constant threats, spent two decades reporting on crime and corruption near Colombia’s border with Venezuela. The post Cristian Herrera reported where fear tried to silence others. Then he was killed appeared first on LatAm Journalism Review by the Knight Center .
Alumni college leaders explore the future of higher ed
The panelists considered three key issues facing colleges and universities: rapidly advancing technology; an altered relationship with the federal government; and an erosion of public trust in higher education. The post Alumni college leaders explore the future of higher ed appeared first on Cornell AI Initiative .
Amazon partnership establishes Cornell AI security initiative
Cornell computer scientists will lead the development of safety protocols to shore up AI agents and the code they produce. The post Amazon partnership establishes Cornell AI security initiative appeared first on Cornell AI Initiative .
😸 WATCH: Sleeping on Microsoft AI? Whoops.
Microsoft Build sit-down with Mustafa Suleyman on Microsoft’s 7-model AI flex
¡Trabaja en CENIA!
El Centro Nacional de Inteligencia Artificial (CENIA) es una corporación público-privada, financiada por la Agencia Nacional de Investigación y Desarrollo (ANID) desde noviembre de 2021. Nuestro propósito es poner la Inteligencia Artificial al servicio de las personas y transformar a Chile en un referente en el desarrollo científico y tecnológico de la IA en Latinoamérica […] The post ¡Trabaja en CENIA! appeared first on CENIA .
Metric Semantic Layer: How Lyft Governs and Scales Key Data Definitions
Written by Rohit Channe and Simran Mirchandani at Lyft. Motivation At Lyft, data isn’t just a resource — it’s woven into everything we do. Metrics drive key forecasts, steer operational decisions, and put our boldest hypotheses to the test. But as Lyft scaled, products launched and evolved, and team members came and went, we found ourselves at risk of different teams using different definitions for a given metric. What did “Metric ABC” actually mean? The answer often depended on the context and application of the team you asked. The consequences were predictable. Without centralized version control or a shared standard, outdated metric definitions crept into decision-making. Our solution was to build an internal Metric Semantic Layer (MSL) : a centralized repository that serves as a single, authoritative home for every metric’s definition — providing both a clear, plain-English description and the definitive SQL code. No more hunting across codebases or tribal knowledge — just one place to store and access a standardized, agreed-upon definition. With MSL, we have a single source of truth — consistent terminology and assumptions across every team, so everyone is genuinely speaking the same language. We achieve this through three key principles: Simplified onboarding and change management — update a metric definition once, and the change automatically and frictionlessly flows through every downstream application that depends on it Intentional governance — clarified ownership,…
Campanhas eleitorais inteligentes: Data Privacy Brasil contribui para trilha do RenovaBR sobre segurança digital
A Data Privacy Brasil participou da trilha online sobre segurança digital em campanhas eleitorais promovida pelo RenovaBR, iniciativa voltada à formação de lideranças políticas e equipes de campanha para os desafios do ambiente digital contemporâneo. A atividade reuniu especialistas para discutir riscos, responsabilidades e boas práticas relacionadas ao uso de tecnologias, dados pessoais e inteligência artificial durante os processos eleitorais. O post Campanhas eleitorais inteligentes: Data Privacy Brasil contribui para trilha do RenovaBR sobre segurança digital apareceu primeiro em Data Privacy Brasil Research .
Portable vLLM Model Inference Kernels in Helion
TL;DR Helion kernels were integrated into vLLM for FP8 inference using Qwen3 models and evaluated across NVIDIA H100 and B200 GPUs. The experiments show that Helion provides a productive PyTorch-native...
Give GitHub Copilot CLI real code intelligence with language servers
Install and configure LSP servers for GitHub Copilot CLI, replacing brute-force grep/decompile with real code intelligence. The post Give GitHub Copilot CLI real code intelligence with language servers appeared first on The GitHub Blog .
Report and Community Resources: Corporate Power Players in the Data Center Industry
This working draft of AI Now’s upcoming report traces corporate power in the data center industry in the United States, focusing on the flows of money and power that determine who both drives and benefits from the current data center boom. The aim of this research is to help local communities and their advocates fight […] The post Report and Community Resources: Corporate Power Players in the Data Center Industry appeared first on AI Now Institute .
😸 Claude Fable Five is Anthropic's Most Controversial Model Yet
PLUS: New Claude Models Fables 5 and Mythos 5, Explained
Hubris and the Fog of War: Unpacking the U.S.-Iran Conflict
Millions of Iranians oppose their government, but that doesn't mean they want foreign bombs. Neda Bolourchi explores the ethical gray zones of this conflict.
Announcing Stack Overflow for Agents…
If your coding agent has questions, Stack Overflow for Agents has answers, now in beta.
The Gap Between the Press Release and the Power Grid
Back in February, I wrote about what I called the “Data Center Rebellion,” the growing local resistance to the physical infrastructure behind AI. Since then, I have been asking tech people around the Bay Area how closely they are following the backlash. The answer is usually: they know it exists, but not much more than Continue reading "The Gap Between the Press Release and the Power Grid" The post The Gap Between the Press Release and the Power Grid appeared first on Gradient Flow .
Encoding Your Domain Expert: The Context Layer Behind Spotify's Data Assistant
At Spotify, data problems used to follow a specific pattern. You'd look for the relevant dashboard, there... The post Encoding Your Domain Expert: The Context Layer Behind Spotify's Data Assistant appeared first on Spotify Engineering .
Fully autonomous drones have killed human soldiers for the first time
A senior figure in the Ukrainian defence industry told New Scientist that a test took place two years ago involving fully autonomous drones set to destroy anything in a given area, with confirmed casualties
Timing Trick Cuts Energy Used in LLM Training by Up to 14 Percent
OpenAI ’s fourth large language model (LLM), GPT-4 , took an estimated 50 gigawatt-hours to train, or the equivalent of 5,000 American homes ’ yearly power consumption. That was in 2023. Since then, the computational resources used to train frontier LLMs have only increased , though direct power usage numbers are hard to come by. Now, a research group at the University of Twente in the Netherlands has shown that you can save up to 14 percent of the energy used in LLM training without sacrificing speed by cleverly adjusting the clock frequency of the GPU during computation. Jeffrey Spaan , Ph.D. candidate at University of Twente and lead author on the article, presented the results at the Computing Frontiers conference in Catania, Sicily, last month. “My research is about finding computing waste,” Spaan says. “It’s similar to underutilization of the hardware, but instead of optimizing the software for the hardware, we try to optimize the hardware for the software.” Making the GPU tick Spaan and his collaborators accomplished this by using a technique known as dynamic voltage and frequency scaling ( DVFS ). Every chip—including the GPUs commonly used for training frontier models—uses at least one clock to orchestrate computations. Each operation in the chip is triggered by a clock pulse. The frequency with which that clock ticks controls how fast the chip operates and how much power it draws. Modern GPUs have two clocks, one for the computational core and one for the memory. W…
Paul Kennedy on Great Powers
What a profound honor to have Paul Kennedy on the ChinaTalk podcast.
A Waymo nearly hit me, but I’m still optimistic about driverless cars
A near miss with a Waymo while cycling through London hasn't changed my optimistic stance on driverless cars, but we can't ever let our guard down, says Matthew Sparkes
How to use virtual environments in Python
Of all the reasons Python is a hit with developers, one of the biggest is its broad and ever-expanding selection of third-party packages. Convenient toolkits for everything from ingesting and formatting data to high-speed math and machine learning are just an import or pip install away. But what happens when those packages don’t play nice with each other? What do you do when different Python projects need competing or incompatible versions of the same add-ons? That’s where Python virtual environments come into play. What are Python virtual environments? A virtual environment is a way to have multiple, parallel instances of the Python interpreter, each with different sets of packages and different configurations. Each virtual environment contains a discrete copy of the Python interpreter, including copies of its support utilities (such as the package manager pip). The packages installed in each virtual environment are seen only in that virtual environment and no other. Even large, complex packages with platform-dependent binaries can be corralled off from each other in virtual environments. Why use Python virtual environments? There are a few common use cases for a virtual environment: You’re developing multiple projects that depend on different versions of the same packages, or you have a project that must be isolated from certain packages because of a namespace collision. This is the most standard use case. You’re working in a Python environment where you can’t modify the s…
Rwanda creates dedicated National AI Agency
President Kagame approves formation of Rwanda’s dedicated AI institution
Will a 80 GB GPU and a 48 GB GPU give identical results on an open source text-to-video model for the same quantization and seed?
I am considering to buy GPUs for my project of open source text-to-video models like ltx-2-19b (lightricks) or wan-v2.2-a14b. I read online that the same configuration/quantization and seed will give similar results in quality, only difference is in speed/latency of generation. Is this true? Or will there be a difference ?
[AINews] Anthropic Claude Fable 5 — Mythos but Safe, with Controversial Terms
The much anticipated launch of the Mythos-class model was marred by some controversial usage policies
Agentic AI Governance: Designing for Accountability and Control
Many organizations are already deploying agentic workflows. Some are still experimental, while others are running in production. Once an AI agent can take action on behalf of a business, the question is no longer whether it’s useful, but what happens when something goes wrong. It’s tempting to focus on blame: the AI vendor, the manager, […]
I am trying to understand Monotonic Alignment Search
I have already come into the part of CTC, and I am reading the paper of Flow-TTS recently. What I cannot understand is that the algorithm did not rely on the label, but with the features inside the Mel-spectrogram, then it can alignment the tokens. I know it is trying to predict the probability each frame is by the token, but I cannot quite understand the loss part. Seems it is combined with encoded mel-spectrogram, mu and sigma. And I am confused about that. Thank you very much..
Claude Fable 5 and new AI safety fables
One step further into the power politics of frontier AI systems.
Is RAG Dead? Lessons from Building AI for Tax Law with Alex Bowcut - #769
As context windows grow into the millions of tokens, many AI practitioners are questioning whether retrieval-augmented generation (RAG) is still necessary. If modern models can ingest entire libraries of documents, why bother with retrieval at all? In this episode, Alex Bowcut, Head of Engineering at Sphere, explains why the answer depends on the application. Sphere uses AI to automate global tax compliance—an environment where getting the answer right isn’t enough. Every conclusion must be backed by the correct legal citation, and every decision must withstand expert review. We explore how Sphere built TRAM (Tax Review and Assessment Model), a production AI system that combines retrieval, reasoning models, legal review workflows, reinforcement learning, and deterministic systems to help tax experts move nearly two orders of magnitude faster while maintaining accuracy. Along the way, we discuss why RAG remains critical in high-stakes domains, how Sphere processes legal and regulatory documents from jurisdictions around the world, retrieval architectures, semantic chunking, dense versus sparse retrieval, expert feedback loops, and the challenges of building AI systems that people can actually trust. 🗒️ Full show notes: https://twimlai.com/go/769.
From Chaos to Clarity: How We Built a Unified, Self-Routing Support Ops Ticketing System at Lyft
Written by Atul Gupta , Analytics Manager — LUS Support Ops, Lyft At Lyft, getting operators and riders connected quickly and reliably depends on more than technology — it depends on the teams working behind the scenes to keep that technology running smoothly. For the operators managing Lyft’s fleet across markets, having fast, reliable access to support is what keeps bikes on the road, stations stocked, and issues resolved before they affect riders. Building the infrastructure that makes that support possible is what our team does; this is the story of how we built it. When I first joined Lyft Urban Solutions’ (LUS) Support Ops team in 2020, ticketing processes for our operators were still being established. There was no reliable way to raise issues, track progress, or get routed to the right person. We had a Jira Help Center, but it had become increasingly difficult to navigate. What followed was a five-year journey of transforming that chaos into a streamlined, automated, self-routing system that now handles thousands of tickets per year, with one third of those routed automatically — saving hours of manual triage work annually. The Problem: Organic Growth Gone Wrong On the surface, a Jira Help Center sounds like a reasonable solution. In practice, ours had become a maze. Here’s what we were dealing with: Duplicate intake forms doing the same job under different names Redundant categories with no clear ownership Forms that didn’t capture the right information upfront, for…
What it feels like to work with Mythos
Claude Fable represents another big jump in AI
Can robots read the room?
Artificial intelligence may one day give robots social intelligence, but so far, existing models do a poor job of using human facial cues to predict the outcome of a situation. The post Can robots read the room? appeared first on Cornell AI Initiative .
What Do AI Standards Mean for Small and Medium Enterprises?
While AI standards and best practices provide valuable guidance to practitioners, they often are geared toward integrating AI into the structure and practices of large, well-resourced organizations. Yet small and medium enterprises (SMEs) stand to benefit greatly from AI adoption as well. This blog examines the implications of AI standards for smaller organizations and proposes several achievable initial steps that practitioners can take to further responsible AI deployment under resource constraints. The post What Do AI Standards Mean for Small and Medium Enterprises? appeared first on Center for Security and Emerging Technology .
Vector welcomes Canada’s AI Strategy: AI for All
Prime Minister Carney launched AI for All, Canada’s new national AI strategy. It’s a moment that Vector, and anyone dedicated to building Canada’s AI future, have been diligently working towards. […] The post Vector welcomes Canada’s AI Strategy: AI for All appeared first on Vector Institute for Artificial Intelligence .
Why don’t cancer medicines work the same for everyone?
The post Why don’t cancer medicines work the same for everyone? appeared first on Source .
Dadocracia – Ep. 201 – IA nos Apps de Relacionamento
No episódio 201 do Dadocracia, falamos sobre as mudanças nos apps de relacionamento: recursos de auxílio para conversas, seleção de fotos e, em alguns casos, a preferência de parte dos usuários por se relacionar diretamente com sistemas de IA. O post Dadocracia – Ep. 201 – IA nos Apps de Relacionamento apareceu primeiro em Data Privacy Brasil Research .
The patch model is breaking. AI evaluation needs a new way to disclose what it finds.
- Why traditional vulnerability disclosure fails for open-weight models—and how we are building a new standard for AI evaluation. The post The patch model is breaking. AI evaluation needs a new way to disclose what it finds. appeared first on MLCommons .
Hey Siri, meet AI
what's the deal with loops