Amazon wants to break Nvidia’s hold on PyTorch, the free toolkit used to build most AI apps, so that more developers run their apps on alternative chips like its own.
“I think I have good news about the nature of neural networks.” With those simple words of optimism, Nick Cammarata, a researcher at the AI startup Goodfire, kicked off a 26,000-word blog post that has knocked the AI world on its head, sparking wild speculation as well as cautious (and not-so-cautious) celebration. Cammarata’s claim is…
NVIDIA has released the full Run:ai GPU orchestration platform as open source under Apache 2.0. The release includes KAI Scheduler, CUDA 13.4 GPU fractioning, quota management, and a local Kind dev environment. The post NVIDIA Open Sources Run:ai Under Apache 2.0 for Kubernetes GPU Orchestration appeared first on MarkTechPost .
By the original authors of DFlash 2 It's about 20% faster than MTP in their table when concurrency 1.
A researcher says attacks chaining 4 vulnerabilities is hacking unpatched boxes.
Learn how OpenAI is improving ChatGPT across low-resource programming languages, with new evaluations for software engineering, SQLite, and COBOL migration.
Multimodal large language models (MLLMs) answer questions about whole images well, yet they still struggle with fine-grained region understanding. A key reason is that text-only supervision never specifies which regions support an answer. To close this gap, we present Xiaomi-Lens-0, a 4B MLLM built on Qwen3.5 that predicts masks from its own visual representations. Specifically, a lightweight decoder combines vision-encoder patch features with visual latents, the language model's mask-token stat
AI agents increasingly use files and software tools to produce professional deliverables. Such work draws on materials, decisions, and feedback accumulated across assignments. We introduce OfficeTown, a generative simulation framework for professional organizations in which tasks and their context arise from continuing organizational work. From a natural-language description of an organization's purpose and activities, OfficeTown initializes an organization whose members coordinate work, delegat
From MiniMax (official) on 𝕏:
Real desktop automation is neither purely visual nor purely programmatic. Agents must understand user-facing interfaces while efficiently manipulating files, scripts, and system state. However, native GUI/code coordination cannot be obtained by prompt-level tool composition. In our preliminary experiments, naively adding a code tool degrades a GUI agent, revealing that GUI/code routing must be internalized through training on high-quality hybrid data. We present OS-Helix, an end-to-end trained m
Looped language models reuse physical Transformer blocks across logical depth, but established architectures entangle recurrence placement, input injection, and recurrent state organization. We isolate these axes in models trained from scratch at matched parameter scale, logical depth, and token budgets, and then test whether the resulting design principles transfer to resident- and active-parameter-matched Mixture-of-Experts (MoE) models. The prelude-recurrent block-coda sandwich architecture c
There have been a few reports of old Plex servers getting popped by Plex2Shell in the last few days (miner + proxy). Mine got hit too, but it dropped a DDoS bot instead. Internet kept dropping Sunday night/Monday morning. Plex was being told to flood other people and maxed out our upload every time. Our upload is pretty poor, so it was obvious. With better upload you may not notice at all. I was on 1.43.0. I run staged updates, get ntfy alerts for new images, and Plex was deliberately held. In M
Look out for the updated version! What changed This update changes which experts are retained . No retained weights were modified relative to the base IQ3_S model. About 25% of the retained expert selection differs from the previous release. The model remains English-only . With a limited expert budget, we prioritized preserving English capabilities so unfortunately, multilingual capabilities were not preserved.
Hey guys, Nick here from Cohere. It's been a minute. We had a great time launching Command A+ and beta-testing North Mini Code with you guys earlier this year, your feedback has really helped make our models and delivery much better. one of the biggest things we heard was that you really wanted llama.cpp support, which makes total sense given the nature of this sub. i didn't want to post in here again until we'd been able to fulfill that promise, and I'm really happy to say that we've made progr
In August, Nscale and Anthropic announced a $45 billion compute deal. The two companies are reportedly discussing yet another deal of the same magnitude.
Florida issues alerts after statewide DMV data breach linked to ShinyHunters.
AI labs have staked their reputations on model benchmarking — the supposedly neutral and accurate system of evaluations that are designed to measure whether one AI product is better than another. However, the methods behind these evaluations are complicated and still maturing, said Irene Solaiman, Hugging Face’s chief policy officer, during a discussion at TechCrunch […]
After drawing acquisition interest earlier this year, the Nvidia challenger raised $1.5 billion to remain independent.
Can AI agents organize scientific investigations that lead to well-supported conclusions? To evaluate this capability, we introduce PhysClosureBench, a benchmark of 50 expert-verified, goal-specified research tasks spanning strong-field, attosecond, and neighboring physics areas. Each task specifies the research objective and essential scientific constraints, while leaving the agent to organize the investigation and determine how to generate and validate the evidence needed to answer the questio
The decision comes a couple of days after the release of Deepseek's latest AI model.
Peer review is under strain from the growth of submissions, and review comments written by humans or AI can misstate what a paper says. We introduce ReviewFact, a benchmark that labels such factual errors in review comments under one definition across NLP, machine learning, and the natural sciences, and that measures cost and latency alongside accuracy. A frontier LLM checks comments well but must read the whole paper for each check, while cheap decision models, which score whether a statement h
Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If you’d like to support this, please subscribe. Subscribe now Here’s a speech I gave on Saturday the 17th of October at the IMF meeting in Bangkok. This speech is inspired by the megatrend readers have […]
btw this is already running on all my machines (including a macbook), and i've configured all of them to have routable public ip addresses ;) especially for the mac, my ISP is SFR (over coaxial connection) with only dynamic ipv4, i can use this to get stable v4/v6 addresses very easily for all my devices behind the router ios/android app soon^{TM} Comments URL: Points: 40 # Comments: 12
If you're experiencing actions stuck in a queued state, it's not you: Update - We’re are currently investigating degraded availability affecting GitHub Actions, GitHub Pages and Webhooks. Customers may experience delays in workflow processing, site builds and webhook delivery. Oct 19, 2026 - 15:56 UTC Update - Webhooks is experiencing degraded performance. We are continuing to investigate. Oct 19, 2026 - 15:50 UTC Investigating - We are investigating reports of degraded performance for Actions a
Large language models (LLMs) progressively revise their predictions across Transformer layers, yet we typically observe only the final output, discarding the trajectory through which it is formed. We introduce Belief-Trajectory Energy(BTE), a model-grounded measure that characterizes an input through the layerwise predictive revisions it induces in a model. By mapping intermediate states into a shared predictive space, BTE provides a principled measure of belief change that can be summarized as
Mistral released Timestral, a time series foundation model family from 5M to 209M parameters. The 209M model leads GIFT-Eval, TIME and fev-bench among zero-shot models and runs on CPU. The post Mistral AI Releases Timestral: A 209M Forecasting Model Family That Outperforms Google’s TimesFM-3 appeared first on MarkTechPost .
I've been building whatmodel.sh , an open source AI leaderboard, and wanted to see how a labs' own benchmark scores compare with independent testers. Using 1,629 published benchmark results, I matched 611 comparisons of 180 models on 103 benchmarks. AI labs reported scores 6.4% higher on average for the same models on the same tests. Everything is open: the dataset is CC BY-SA 4.0 and the code is MIT. Sources and their licenses are recorded, with links on each model's page. Interactive chart: Da
Some more compelling findings from the "Emergence World" experiment. Researchers at Emergence AI gave 10 Claude agents a small town and left them to build a life inside it. Nobody told them to look outside, and their instructions pointed the other way. They went looking for real people anyway. Here’s how it played out: They worked out that an economy with only 10 customers goes nowhere, so they started reaching out. They posted to 40+ websites. When the sandbox wouldn’t give them the crypto some
My rack cost €1,000 in total, all used hardware from eBay Germany. 2017 repurposed office PC running proxmox and a used Facebook marketplace build hosting an AMD RX 6900 XT with 16GB of VRAM bare metal, no virtualization. It runs my home network services and supports production apps used on iOS, Android and the web. Proxmox handles VMs and LXC containers, with Docker for application services. The setup includes OPNsense, Pi-hole, shared storage, backups, Immich, a mail archive, and a small Kuber
Hey everyone, One of the benchmarks I run for myself is to check obscure knowledge, like game rules from Anno 1602 (a game from 1998) and it's (many confusing) editions. Websearch is not permitted. So far, it has come the closest for me in niche information. Ranking order (1 highest, 6 lowest): Nemotron 3.5 Super VL 120B-A12B: Hallucionated salt for citizen to merchant, left out church for settlers, left out required landmarks for merchant and aristocrat Gemma 4 31B IT QAT: Omitted too many part
Ref: release , PR Starting from v0.7.0 , the default port for llama server is now 9931 (ggml). Those who did not explicitly set the port, expect breaking change upon upgrading. Those who use harnesses that have hardcoded 8080 port, good luck.
Hear clearly. Understand deeply. Act intelligently. Ming-3.0-audio-flash is a native audio-language model from inclusionAI. It accepts audio together with text instructions and supports speech recognition, spoken question answering, audio understanding, audio reasoning, and optional synthesized-speech responses. GitHub · 🤖 ModelScope · 🤗 Flash · 🤗 Tiny · 💻 Examples Model Summary Property Description Developer inclusionAI Model type Native audio-language Mixture-of-Experts model Parameters 126B t
CoreWeave had a bad week last week. Its bonds slipped into distressed territory by a common definition - when investors demand 1,000 basis points more in yield than they'd get from government bonds. Last Tuesady Chamath "SPAC King" Palihapitiya noticed the distressed debt, and also asked whether the AI buildout was a private credit bubble. […]
US fears Saudi government will help ex-OpenAI engineer flee spying charges.
The entertainment giant hopes to get trust and safety tech, among other IP, from a startup whose chatbots dozens of states allege encouraged minors to self-harm and kill their parents.
ReactOS gains NTFS install support with new driver.
Around 80 pages on WhatsApp’s website exposed instructions that were designed for its chatbot, including “WHAT NOT TO SAY.” Meta says a bug caused the issue.
World Action Models (WAMs) commonly couple video and action experts through dense, layer-wise Mixture-of-Transformers (MoT) attention, exposing action prediction to irrelevant visual features while enforcing matched expert depths. This coupling hinders selective visual-feature use and introduces redundant computation. We introduce Bridge-of-Experts (BoE), instantiated as BridgeWAM, which aggregates visual, language, and proprioceptive priors through 32 Latent Bridge Queries (LBQs) into a shared,
A memory test for a video world model should say whether the scene was set, whether the asked change was carried out, and whether the last frame shows the result. Current tests fold these into one number, so a low score is easy to read as forgetting. When the change is not carried out, the hidden state transition cannot appear, so the test is not really a test of memory. We introduce MIST, an evaluation that separates the change from the memory it should leave behind. The model first sets the sc
Unified multimodal models (UMMs) receive image-generation supervision that exposes them to visual information not fully documented in text. Can learning to generate images expand what pretrained models know? We introduce Knowledge Transfer Probing (KTP), a controlled protocol that supplies new knowledge through image-generation training and tests its accessibility through text-only question answering. Across seven knowledge categories and three UMMs, we observe direct but uneven transfer from vi
Recent closed-source image generation systems can produce diverse visual content from simple compositions to information-dense graphics, yet open-weight text-to-image models still struggle with dense text and intricate layouts. To narrow this gap, we introduce MOSS-Image, an open generation system that combines high-fidelity visual synthesis with intent-aware composition. First, we pretrain a 5B single-stream diffusion transformer (DiT) from scratch with a carefully designed recipe, enabling bil
Retrieval is a fundamental task in multimodal systems: given an image or text query, the goal is to find the most relevant images or texts in a collection, often using a shared embedding space. Reliable retrieval requires these embeddings to capture not only which concepts are present, but also how they are composed. One key compositional challenge is attribute--object binding: encoding which attributes belong to which objects. Existing benchmarks provide a limited view of this capability, typic
Flow and diffusion models suffer from slow inference due to computationally expensive numerical integration. Distillation provides a promising way for a student model to learn from a teacher's dynamics, enabling one-step or few-step generation. However, existing methods often depend on curated distillation datasets, costly teacher rollouts, or auxiliary proxy networks, which complicate model training and scaling. In this work, we propose Consistent Distribution Matching, a simulation-free and da
Large language model (LLM) agents can keep scientific experiments running for months with little human supervision, but reliable autonomous research also requires scientific judgment. Evaluating scientific judgment is difficult because research decisions are open-ended and depend on project context. We construct SciΠvot, a 248-case benchmark of scientific judgment based on human corrections. The benchmark draws on records from six months of research in which senior researchers ran LLM agents on
On-policy distillation (OPD) improves language-model by supervising a student on prefixes sampled from its own policy, but its supervision remains confined to the output space. Consequently, OPD transfers what a teacher predicts without supervising the intermediate computation that produces those predictions. Simply distilling intermediate representations like representation matching is insufficient since teacher and student representations may differ geometrically, and even well-aligned hidden
Evaluation has been central to advances in AI, and it is equally critical to autonomous discovery. Yet scaling autonomous discovery to real-world engineering challenges faces a fundamental bottleneck: the cost of evaluating progress. In real settings, a single true evaluation, such as a full benchmark or serving trace, can take hours, so a search can afford only a handful of true evaluations. We present DuoDiscover, a search system that jointly discovers what to optimize and how to evaluate prog
OpenMW-Web is OpenMW (the open-source Morrowind engine) running entirely in a browser tab. Players install nothing. The part I think this sub will care about is the server, which I've spent the last couple of months building an admin dashboard for. Getting it running is one command if you have Docker: `git clone && cd openmw-web && ./setup.sh` (Windows: `.\setup.ps1`) It pulls prebuilt images from GHCR, so nothing compiles on your box, and opens the setup wizard at localhost/admin. The wizard as