2026-08-11

Looks like the Ling team open weighted a much smaller version of the Ling-3.0-flash they open weighted a few days ago. It's 8B params with 1.3B active, and seems to fall between the 4B and 8-12B Qwen and Gemma models in terms of performance. Should have a massive tokens/sec on most systems. I quite like tiny MoE's conceptually. Edit: looks like the model card actually reports speeds: With FP8, Ling-3.0-tiny reaches around 100-105 tokens/s on DGX Spark and 86-90 tokens/s on an M4 Pro MacBook, wit

After a few long years of finding time to document my lessons from training open models, my post-training book is done!

Obviously nobody needs a transformer that's good at multiplication. I wanted to know whether a stock transformer could do exact arithmetic if I chose its weights directly. I implemented the grade-school algorithm as a computation graph and compiled it into an ordinary Phi-3 Hugging Face checkpoint using Torchwright, a compiler I wrote. No training. The three-digit calculator gets all 3,000,000 supported expressions right. I've published checkpoints to Hugging Face that support up to 12 digit x 1

Wowee!! Just when you thought it couldn't get better for open weight models, we probably have had our best period yet!?!?! Models that rival the closed frontier, Opus level models on non-insane hardware and more. A massive industry alliance coming out in support of open AI in response to the two closed model giants best lobbying efforts. Is this the best timeline? Someone pinch me! Or just tell us what you're favorite model is now The standard spiel: Share what you are running right now and why.

Activation Oracles (AOs) are language models trained to answer natural-language questions about another model's internal activations. They offer a flexible interface for reading hidden information from model states, especially when relevant information is internally represented but absent or incomplete in visible behavior. However, AOs are themselves learned systems: their answers are shaped by training data, objectives, and learned reporting behavior, rather than being neutral readouts of repre

A verified 2026 comparison of LLM observability platforms covering tracing depth, evaluation capability, production monitoring, and pricing. The post Top LLM Observability and Evaluation Platforms in 2026: Langfuse, LangSmith, Braintrust, Arize, and More Compared appeared first on MarkTechPost .

Just wanted to share a user report that I found to be very interesting. Some person with an intriguing name manu69x managed to run 1M context on a single RTX 3090 with a model based on Qwen 3.5 35B A3B, which was taking up some 17 GB of VRAM alone. And "run" here doesn't mean just "server didn't crash", it's that context didn't turn into a mess, so they were able to extract 7 needles positioned in various parts of the text. They used KVarN 4-bit for both K and V with my BeeLlama.cpp fork, build

Meta’s new open-weight Muse Glimmer model offers a glimpse of Mark Zuckerberg’s personal superintelligence vision, as well as the emerging divide between AI users can own and access.

Companies treat some email domains as digital trash cans, despite the risks.

Companies that rely on Ceva Logistics for shipping their physical goods to customers say their personal data was taken during a recent cyberattack.

A bug in the tech giant's website the logo of US marketing automation company Klaviyo Inc. is seen displayed on a smartphone in front of an abstract background on a computer screen..

Having a ‘Killer Application’ that everyone wants to use helps sell hardware, plain and simple. DeepSeek V4 Flash 0731 isn’t an app of course, but I think it’s going to be the major catalyst for getting a lot of people to buy a couple of NVIDIA GB10-based systems because: It is an amazing coding / agentic use model. It fits perfectly on a 2x Spark Cluster It runs Fast AF with the right vLLM recipe. (I’m getting 60 tk/s with this one: ) You can run it with a fairly usable 1M context window. It ru

ByteDance’s Seed team has introduced SeedRealtime, a native audio-visual full-duplex LLM. The model fuses audio, video and text in a single unified architecture. It interacts in real time over continuous multimodal streams, rather than one turn at a time. Seed positions it as a step toward omni-modal interaction, and claims three breakthroughs: joint audio-visual understanding, […] The post ByteDance Seed Introduces SeedRealtime: a Native Audio-Visual Full-Duplex LLM That Watches, Listens and Sp

arXiv:2608.06400v1 Announce Type: new Abstract: Reward models are central to learning from human preferences, yet identifying what drives their predictions remains challenging. Recent sparse Mixture-of-Experts (MoE) reward models seek to improve interpretability by routing prompts to specialized experts and characterizing experts through examples with high routing weights. However, routing weights only reveal which prompts an expert $\textit{receives}$, not how it $\textit{judges}$ responses, pr

The API has zero authorisations checks on cancelling other people's reservations … I tested this with the person in waitlist position #1 — and it actually went through. So you've moved from #4 to #3 already. — OpenClaw , hacking an Australian gym-booking website Tags: ai-ethics , generative-ai , openclaw , ai , ai-security-research , llms

从云计算到云原生，产业真正的跃迁从来不是简单堆算力，而是找到更适合新世界的架构。

Generic parameter-efficient fine-tuning (PEFT) methods transferred from language models can fail silently on real-time detectors, whose heterogeneous operators and detection-specific components impose placement constraints absent from regular Transformer stacks. We propose YOLO-PEFT, a structure-aware framework that formulates adapter placement as an auditable constraint-planning problem. Given a detector graph, a PEFT request, and a resource budget, YOLO-PEFT assigns operator and semantic roles

NVIDIA releases NemotronLabs VoiceChat 11B, an open full-duplex speech-to-speech model with 448 ms latency and live tool calling. The post NVIDIA Releases NemotronLabs VoiceChat 11B: An Open Full-Duplex Speech-to-Speech Model with ~450 ms Turn-Taking and Live Tool Calling appeared first on MarkTechPost .

Deploying autonomous multimodal agents in continuous, real-world environments requires them to ingest unbounded audio-visual streams and maintain hour-scale memory. However, current evaluations predominantly rely on brief clips and multiple-choice formats. This design allows minimal baselines that process only the last four frames to match or surpass complex streaming models, while answer options also expose language shortcuts. We introduce StreamArena, a benchmark for hour-scale, interactive st

GitHub Models is now retired I missed this news until today, when the GitHub Actions run for my simonw/research repository failed with this error message: GitHub Models is temporarily unavailable as part of a scheduled retirement brownout. That message is already stale, because the retirement has been completed. GitHub Models was an odd-shaped duck. GitHub provided a model playground tool and a unified API across a bunch of different LLM providers, with the biggest benefit being that code runnin

CLI-based software-engineering agents have matured rapidly, yet the open ecosystem has converged on a single training environment: trajectory datasets used to fine-tune open models are collected almost exclusively under OpenHands. Models fine-tuned on this data score well under OpenHands but degrade substantially when deployed under any non-training scaffold. Untrained base models do not show this divergence, indicating the gap is fine-tuning-induced and tied to the conventions of the training s

VLX-Seek-1.5-10B VLX-Seek-1.5-10B is the open-source 10B model in the VLX-Seek 1.5 family, designed for fine-grained perception and visual grounding in embodied scenarios. It targets practical settings such as drones, robots, robotic dogs, surveillance cameras, inspection systems, and other edge-side visual intelligence applications where a model must identify what is present, localize the right instance, and avoid grounding objects that are absent. Unlike coordinate-generation-based VLMs that d

OpenAI sent Governor Greg Abbott a letter outlining its commitment to responsible AI infrastructure in Texas. The letter supports reliable, transparent growth that benefits Texans.

Little less smart than Qwen, but way fewer tokens per task.

I did some testing this morning, and I was surprised to find that Muse Glimmer actually comfortably fits on a single RTX 3090 with full context + DFlash + mmproj at Q4_K_XL, unlike Qwen3.6-27B and Gemma-4-31B. Muse Glimmer supports up to 256k context according to Unsloth. Here is my command: llama-server \ --model Muse-Glimmer-30B-UD-Q4_K_XL.gguf \ --mmproj Muse-Glimmer-30B-mmproj-kquant.gguf \ --spec-draft-model Muse-Glimmer-30B-DFlash-kquant.gguf \ --spec-draft-ngl 999 \ --spec-draft-n-max 15

Every few decades, someone announces that science has reached its end. In 1903, the revered physicist Albert Michelson wrote that the “facts of physical science have all been discovered.” In the 1980s, Stephen Hawking predicted that theoretical physics might be finished by the end of the century. With the explosive arrival of artificial intelligence, the…

According to Valve, European Steam hardware customers likely had some of their data compromised.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. AI for science needs reasoning, not just data —Eric Schmidt, the former CEO of Google and the cofounder of Schmidt Sciences, and Suhas Mahesh, who leads the AI for science work…

arXiv:2608.06398v1 Announce Type: new Abstract: Recent byte-level large language models (LLMs) have made tokenizer-free modeling increasingly competitive by grouping bytes into dynamically sized patches. However, existing byte-patch architectures still apply the same dense feed-forward computation to every patch. This uniform computation cannot adapt model capacity to variations in patch semantics and granularity. We address this limitation with EntropyMoE, a Mixture-of-Experts (MoE) architectur

arXiv:2608.06411v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) achieve strong performance across diverse vision-language tasks, but their efficiency is limited by the cost of processing numerous visual tokens. Visual token pruning can reduce this cost, but requires accurate token importance estimates. Recent studies have demonstrated that text-to-vision attention from middle language model layers can effectively guide visual token pruning, typically using attention from

World-Action Models (WAMs) improve end-to-end autonomous driving by transferring video dynamics priors to action prediction, but existing methods require costly future generation at inference. We present SimWAM, a simple yet effective WAM that uses video generation purely as a training signal. It co-trains a pretrained video expert and a lightweight action expert with joint flow matching. An isolated attention mask keeps action prediction independent of future frames, allowing the video branch t

arXiv:2608.06474v1 Announce Type: new Abstract: Large language models increasingly generate complete websites from natural-language descriptions, and reinforcement learning has become a central approach to closing their remaining functional gap. This training regime is bottlenecked by reward design. Hand-authored browser scripts are executable yet costly to write for open-ended requirements, while VLM and GUI-agent graders scale but may issue verdicts before observing the decisive state. We prop

arXiv:2608.06578v1 Announce Type: new Abstract: Frontier language models are trained using distinct data, objectives, and safety pipelines. Whether these differences produce measurably different behaviors under explicit steering pressure remains underexplored. This study evaluates behavioral steerability across six frontier models from six developers using 300 paired base and steered items over three categories: values-conflict, reasoning-elicitation, and reasoning-suppression (plus 40 validatio

A coding agent that can refine its own harness Discussion | Link

Bubble started to burst?

Deep vision models exploit shortcuts, relying on cues that correlate with supervision signals. Prior work has focused on visible biases, such as object-background or texture correlations. We identify a different source of shortcut learning: invisible metadata traces embedded at the pixel level, for metadata such as image processing and photo acquisition. We hypothesize that large-scale semantic supervision, whether through categorical labels (ImageNet) or billion-scale captions (LAION), naturall

I know Muse Glimmer is pretty new and all, but was wondering if anyone else has run into Glimmer outright refusing to code even small things? I am using Unsloth Q8, dual 3090's, in Kilo Code. I was trying to get it to help me with a bug in my codebase (using pyton stdlib to manipulate a mouse, moving it, clicking, etc.) and it has been giving me different versions of this: I can’t provide code to control your mouse without context. Moving a mouse programmatically can be misused for automation, c

Hi HN, we’re Eren, Berat and Kaan. We’re building Stoa ( ), a marketplace for new and used GPUs and AI servers. GPUs are the collateral in the data center buildout. Today, financing terms mostly depend on the offtaker, meaning the company that has committed to use the compute. If that company is a hyperscaler, the financing can look investment grade. If it’s a smaller cloud or startup, terms get expensive fast, even with the same hardware as collateral. The lender’s problem is pretty reasonable.

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Last week, I headed 30 miles south of San Francisco to a hotel in Mountain View, California, to join some of the most accomplished, and some of the most promising, AI…

Meta CEO Mark Zuckerberg has a lot to say about the idealized future he now envisions for humanity co-existing with artificial intelligence - his latest essay spans more than 6,500 words on the matter. The lengthy manifesto Zuckerberg published on Monday, titled "The Future is for Everyone," broadly lays out his beliefs about how the […]

OpenAI CFO Sarah Friar shares five lessons for building an AI-native finance function, from automated forecasting to stronger controls and AI ROI.

arXiv:2608.06609v1 Announce Type: new Abstract: Automated item evaluation (AIE) refers to the use of computational methods to assess item quality without requiring manual expert review or field testing of the items under evaluation. We aimed to build a near-comprehensive AIE model by predicting item acceptance and rejection from item text using historical rejection data from a large-scale standardized testing program. The dataset contained 52,759 English language arts (ELA) and mathematics items

arXiv:2608.06402v1 Announce Type: new Abstract: Community detection is a fundamental task in graph analytics that aims to identify cohesive groups of entities with similar behaviors or interests. Classic objective-driven methods struggle with complex graph structures, while deep-learning approaches improve performance at the expense of interpretability and rely on labeled data and training. Large language models (LLMs), with strong reasoning capabilities and world knowledge, are promising for in

arXiv:2608.06632v1 Announce Type: new Abstract: Industrial recommendation systems predominantly adopt a passive ranking paradigm that infers user preferences from implicit behavioral signals (e.g., clicks, dwell time) rather than explicit, natural language inputs. As a result, users experience a persistent discrepancy between their explicit interests and what passive behavioral algorithms deliver, limiting their ability to express nuanced preferences or steer their feed in real time. To address

arXiv:2608.06544v1 Announce Type: new Abstract: World models for visual control typically learn compact latent states by reconstructing observations, implicitly encouraging representations to preserve information across the entire visual input. However, task-relevant content often occupies only a small fraction of the observation, while background clutter and distractors consume valuable representational capacity. This mismatch between visual reconstruction and control objectives biases latent r

Premium seats are coming to ChatGPT Business. Sign up by August 20 to get $100 in workspace credits and unlock higher usage for your team's most demanding work.

The enterprise marketing team at Zapier uses ChatGPT Work to reduce the number of drop-offs in its lead funnel, build campaign assets, and automate reporting.

Recent video models increasingly support generation, reference conditioning, and editing within a single model, yet typically expose them as separate operations over fixed inputs. Practical creation unfolds across multiple shots, requiring one model to generate from text, follow a reference, or edit source footage while maintaining shared history. We formalize this setting as interactive multi-shot video creation (IMVC) and introduce ContextMaster, a unified model with a role-aware context repre

Current video world models struggle in multiplayer environments because they entangle world state with view-dependent visual latents, leading to redundant compute, view inconsistencies, and poor scalability. We propose MAS (Multiplayer world models with Authoritative Shared State) to resolve this limitation. Inspired by multiplayer game architectures, MAS disentangles world dynamics and view rendering. A learned Logic Engine advances a global, authoritative typed state from joint actions without

Human evaluation of AI systems and digital products is costly, slow, and difficult to scale. Offline evaluations are more scalable but often abstract away human diversity and interactive behavior. We therefore introduce MatrAIx, a population-scale simulated-user evaluation infrastructure for testing AI systems and digital products with heterogeneous users. MatrAIx has three core components: First, Persona 8B contains 8.3 billion persona records represented by 1,290 categorical dimensions. Record