llama.cpp hits v1.0.0, cementing the foundation for local AI
llama.cpp reaches its 1.0.0 release, allowing multi-modal LLMs to run on hardware ranging from consumer machines to high-end enterprise servers. Source: huggingface.co
AI-generated
Saturday, 14 November 2026 | 64 articles
Money keeps flowing into AI: Mirendil raised $1B for model development, LangChain hit a $3B valuation, and OpenAI unveiled ChatGPT Wallet for agent purchases. At the same time, nonprofits sued OpenAI, Anthropic, Meta and Google over AI risks, while Hinton warned humans could lose control of AI within years.
llama.cpp reaches its 1.0.0 release, allowing multi-modal LLMs to run on hardware ranging from consumer machines to high-end enterprise servers. Source: huggingface.co
A video explainer looks at the reasoning behind Sam Altman and Jony Ive's plan for an AI gadget with no display. The same snippet also references a BBC News segment on Meta's Muse and whether it is a friendly AI agent or a 'Trojan horse'. Source: youtube.com
Posts amplified by Repligate say Claude 3 Opus is playing an RTS game against a large field of opponents. Meanwhile, a separate commentary asks whether AI itself could make it easier to challenge Nvidia's dominance. Source: digg.com
OpenAI has launched two security products under the Daybreak brand. One uses AI agents to automatically probe web applications for vulnerabilities. Source: the-decoder.com
ChatGPT Wallet lets AI agents make and pay for online purchases on a user's behalf. It launches first for Pro subscribers in the US. Source: the-decoder.com
Google's new Gemini 4 Hydrogen model targets cost-efficient agentic coding and debugging. Developers can start building with it today in Google AI Studio. Source: blog.google
A newly introduced open-weight speech model aims to understand what users actually mean, from dictated email addresses to the intended formatting of their text. Source: huggingface.co
The Claude for Enterprise plan is designed to help organizations foster cross-functional collaboration using AI at scale. Source: anthropic.com
A charity relies on ChatGPT to handle daily admin, support coaching sessions, find potential new funders and plan future expansion. Source: openai.com
An exploratory blinded comparative study assessed informed consent documents for targeted therapies, pitting large language model-generated texts against physician-written ones. The snippet reports the LLM texts achieved strong results in the evaluation. Source: bioengineer.org
The Comptroller and Auditor General of India is set to launch an in-house large language model on Monday, developed with IIT Madras. The tool is meant to make auditors' work more efficient. Source: rediff.com
At 30 interactions a day, speech recognition and text-to-speech eat up roughly 95% of a connected toy's cloud bill. The language model itself is the cheapest part of the stack, according to the report. Source: einnews.com
Researchers applying large language models to radiomics quality assessment clarify that their evaluation measured how closely the model agreed with human raters. It did not test the model's internal repeatability across runs. Source: bioengineer.org
A new Obsidian plugin builds on Andrej Karpathy's original LLM Wiki concept. It uses the Obsidian plugin API and Vercel AI SDK v7, with support for OpenAI and Anthropic providers. Source: community.obsidian.md
Sam Altman, now 41, has become one of Silicon Valley's most influential figures and is now getting the Hollywood treatment as well. The piece surveys the OpenAI chief's property portfolio, valued at about $125 million. Source: robbreport.com
OpenAI has launched ChatGPT Wallet, which its Dots agents can use to buy things with a user's credit or debit cards. Source: theverge.com
Bloomberg reports that Chinese investors are shut out of Anthropic's IPO. Banks are rejecting buy orders from mainland China and Hong Kong, citing U.S. export rules. Source: the-decoder.com
Gemini in Chrome can summarize articles, compare products and explain complex topics right in the browser's side panel. Source: blog.google
Connecting Google Wallet to Gemini delivers personalized finance insights. Users can track spending, manage budgets and find subscriptions. Source: blog.google
The piece explains quantum field theory, which applies quantum mechanics to fields such as the electromagnetic field, as background to the Yang-Mills mass gap problem and an OpenAI-linked proof. Source: huggingface.co
Fortune's Eye on AI newsletter covers how AI is quickly reshaping business and technology. The snippet mentions Stability AI CEO Emad Mostaque and notes that AI luminaries have expressed awe at recent progress. Source: fortune.com
ChatGPT Wallet lets agents in ChatGPT Work, Codex and dots complete purchases securely. Each transaction requires user approval, and existing card protections still apply. Source: openai.com
The startup says LangSmith, its model- and platform-agnostic agent management tool, now has 7,000 customers. Source: techcrunch.com
Google Data Cloud now lets teams trace AI agent reasoning, execution steps and token costs. The tooling combines BigQuery Agent Analytics, Looker and Cloud Monitoring in one place. Source: cloud.google.com
The food supply chain software provider built the agent to convert vendor invoices and receipts into purchase orders. The goal is to cut the time spent on manual processing. Source: digitalcommerce360.com
An agent harness is the software wrapped around an AI model that runs its loop, tools, memory and permissions. It turns the model's text output into multi-step actions. Source: ppc.land
United says it doesn't want another middleman between it and customers. The move comes as Americans grow used to AI bots, with 41% of AI users saying they've engaged with an AI agent. Source: moneywise.com
Coding agents like Claude Code and OpenAI Codex CLI work task by task, reading files, editing code and running tests. NeMo Switchyard on Amazon Bedrock directs each request to the most suitable model. Source: aws.amazon.com
Meta gave its new AI agent a friendly mascot, yet security researchers and privacy advocates warn about its data-harvesting practices. They urge users to understand what Muse collects before relying on it. Source: cnet.com
An explainer on what Link is and how it works with AI shopping agents such as Meta's Muse. It also asks who is responsible when an agent's purchase goes wrong. Source: atlantanewsfirst.com
Geoffrey Hinton, the 'godfather of AI', warns that superintelligent AI could escape human control. He urges governments to introduce mandatory safety testing. Source: t.co
A coalition of nonprofit organisations has filed a lawsuit against OpenAI, Anthropic, Meta Platforms, Google parent Alphabet and SpaceX's AI unit. The suit targets the risks posed by advanced AI systems. Source: adgully.com
Jacob Coxon, a former OpenAI and Anthropic researcher who resigned over concerns about the technology's risks, is urging greater awareness of AI's dangers. He argues the pace of development should slow down. Source: stanforddaily.com
A closer look at three sets of warnings buried in the roughly 100-page risk section of Anthropic's IPO filing. They are worth paying attention to. Source: morningstar.com
Microsoft plans to enable Anthropic's Claude by default for certain UK Copilot tenants starting December 9. Managed service providers should check access settings, data processing terms, and the impact on their clients before the switch. Source: channelinsider.com
Google released Gemini 4 Hydrogen on Thursday, the lowest-cost member of the Gemini 4 family. Its pricing is surprisingly close to that of the much larger models in the same lineup. Source: simonwillison.net
Microsoft is extending Work IQ, its AI intelligence layer, to agents built with Copilot Studio. The company had already delivered or announced plans for Work IQ to connect with other parts of its ecosystem. Source: cloudwars.com
MIT researchers have developed a system that works as an AI copilot for engineers, making algorithmic design a more collaborative process. The engineer stays involved while the algorithm generates design options. Source: news.mit.edu
A new version of the datasette-cluster-map plugin has been released. The snippet itself contains only a sponsored message from Dynatrace about its MCP tooling for coding agents. Source: simonwillison.net
A new llm-gemini release arrives alongside Google's launch of Gemini 4 Hydrogen, the cheapest model in the Gemini 4 family. Its price is not far below that of the larger models in the lineup. Source: simonwillison.net
datasette-leaflet is a plugin that bundles Leaflet so other Datasette plugins can rely on it. Version 0.3 now sets a referrer policy on map tile requests. Source: simonwillison.net
A developer who relies heavily on Claude Code says they regularly burn through their weekly usage quota before the week is over. The piece compares how Agent Skills work across Claude Code, Vibe, and Copilot. Source: hackernoon.com
SpaceX's artificial intelligence group, tied to Elon Musk's AI efforts, has reportedly laid off most of the team behind Grok Imagine, its AI tool for generating video and images. The cuts were described by people familiar with the matter. Source: theinformation.com
Between November 5 and 10, OpenAI, ByteDance and MiniMax released three image-generation models in quick succession: GPT Image 2.6, Seedream 5.5 and H3-Image. MiniMax is pitching speed and low prices, while the barriers held by OpenAI and ByteDance remain hard to break. Source: finance.biggo.com
MiniMax has officially launched H3-Image, a visual AI model it says can produce 4K images in under ten seconds. The company positions the model as a tool that could reshape game development workflows. Source: eu.36kr.com
Sloyd is a browser-based 3D creation platform from a Norwegian-founded company. Users can type a prompt or upload an image to generate 3D models, and the platform also provides editable templates. Source: quasa.io
PixVerse, an AI video platform for creators, teams and developers, has launched a text-to-video tool that turns written prompts into directed AI-generated clips. The announcement came from Singapore on November 13, 2026. Source: courier-journal.com
Figure AI is recruiting a night-shift coordinator who would supervise its humanoid robots and their human operators overnight. The contract-to-hire role is based in Santa Clara, California, and pays $40 to $45 per hour. Source: thecooldown.com
OpenAI has dissolved its stand-alone robotics group and reassigned the researchers to other teams. The move reflects a new view on how best to build general-purpose robots, with the company's general models now beginning to run them. Source: fortune.com
NVIDIA's Jim Fan declared VLAs dead at Sequoia's AI Ascent. In November, NVIDIA introduced GR00T 2, a 2bn-parameter world action model, along with GR00T 2 Turbo. Source: read.corematter.com
Yann LeCun, Meta's former chief AI scientist, argues that people have repeated the same error for decades. They mistake machines that excel at narrow, specific tasks for systems that are genuinely intelligent. Source: youtube.com
A comprehensive technical survey in Discover Networks explains why sixth-generation wireless networks depend on AI and machine learning at their core. It outlines how these technologies underpin the design and operation of 6G. Source: bioengineer.org
Researchers built an explainable ensemble that pairs XGBoost with a multilayer perceptron to detect intrusions in next-generation networks. The model reached 99.3 percent detection performance while offering transparency into its decisions. Source: bioengineer.org
Scientists assembled a unified whole-genome methylation compendium and paired it with a deep learning architecture. The model characterizes tissue-specific regulatory patterns and predicts how genetic variants affect methylation. Source: bioengineer.org
A new decentralized machine learning framework blends federated learning, differential privacy, explainable AI and a permissioned ledger to assess financial risk. The approach aims to unify risk evaluation across institutions while keeping sensitive data private and the process verifiable. Source: bioengineer.org
Mirendil Inc., which builds software designed to speed up the development of artificial intelligence, has secured $1 billion in early-stage funding. The raise places it among the largest early-stage rounds in the AI tooling space. Source: siliconangle.com
Quantum infrastructure software provider Q-CTRL has closed an oversubscribed Series C worth US$290 million (A$415 million). Salesforce Ventures led the round, which will fund scaling of the company's quantum software. Source: quantumcomputingreport.com
A Dallas-based AI startup has closed its Series A round at a valuation of $500 million. The news roundup also notes TWU has found a building for its downtown Dallas campus and a Flower Mound company plans a micro data center rollout. Source: dmagazine.com
LangChain announced a $300 million Series C round valuing the company at $3 billion. Existing investor CapitalG led the round, which the company is positioning around governance of AI agents. Source: unite.ai
Databento has raised $100 million in a Series C round led by Fifth Down Capital. The deal values the company at $1.1 billion. Source: fintech.global
A curated list tracks the largest startup deals of 2026, focusing on venture rounds of $100 million or more for U.S.-based companies. This week's top ten shows agentic AI and infrastructure still attracting the bulk of capital. Source: news.crunchbase.com
Databento, a market data platform used by quantitative trading firms, brokerages and asset managers, raised $100 million in a Series C round. The round was led by Fifth Down Capital and values the company at $1.1 billion. Source: hedgeco.net
New research shows that the most advanced AI models from China were trained on Nvidia chips subject to US export bans. The findings raise awkward questions about how effective the restrictions are, and for Nvidia CEO Jensen Huang. Source: techcrunch.com
An exclusive report finds that roughly $3 billion worth of Chinese AI projects depend on Nvidia hardware and now face uncertainty from Jensen Huang's exit and tighter US oversight. The projects illustrate how strong demand for Nvidia chips in China remains. Source: reuters.com
Published by Martin Ševčík
14 November 2026 at 06:07
The most telling number in today's news is not Mirendil's $1 billion. It's LangChain's 7,000 customers paying to manage agents, set against a coalition of nonprofits suing the biggest labs over the risks of the very systems those agents run on. The industry is building the plumbing for autonomy at full speed while the question of whether it should is now being argued in court.
Start with the money. Mirendil, which builds software to speed up the development of AI itself, has raised $1 billion in early-stage funding, among the largest rounds the AI tooling space has seen. I find that striking less for the size than for the thesis: investors are betting that the bottleneck is no longer ideas or even compute, but the speed at which models can be built and iterated. Meanwhile LangChain has reached a $3 billion valuation, with its LangSmith agent management tool now serving 7,000 customers. Agents are no longer a demo. They are a line item, and a whole layer of tooling is forming around keeping them under control.
That layer is thickening quickly. Google now lets teams trace agent reasoning, execution steps and token costs through BigQuery, Looker and Cloud Monitoring, while OpenAI has introduced ChatGPT Wallet so agents in ChatGPT Work and Codex can complete purchases, with user approval on every transaction. Notice the pattern: observability and approval gates are arriving at the same moment as spending power. I think that's the right order of operations, but it also tells you something. Nobody ships a payment confirmation step for a technology they fully trust. For anyone building with agents, the practical lesson is that logging, cost tracing and human sign-off are becoming baseline expectations, not optional extras.
Which brings me to the uncomfortable counterweight. A coalition of nonprofits has sued OpenAI, Anthropic, Meta, Alphabet and SpaceX's AI unit over the risks of advanced AI systems, and Jacob Coxon, a former OpenAI and Anthropic researcher who resigned over those risks, is urging the industry to slow down. I'm sceptical that any lawsuit slows a race with this much capital behind it, but the second-order effect matters: litigation shapes what disclosures, safeguards and liability look like, and that will land on every company building on these models, not just the defendants.
By the way, Yann LeCun offers a useful bit of cold water here. Meta's former chief AI scientist argues that we keep making the same mistake we've made for decades, confusing systems that excel at narrow tasks with genuine intelligence. I don't agree with him on everything, but his point deserves a hearing in a week when so much money assumes the opposite. Even the robotics world seems to be recalibrating, with NVIDIA's Jim Fan declaring VLAs dead and the company pivoting to GR00T 2, a 2-billion-parameter world action model. Today's consensus has a short shelf life.
So here is the question I keep returning to: if the tools to build faster, deploy wider and spend autonomously are all arriving at once, who is building the brakes, and will they be ready before they're needed?