arXiv:2609.05448v1 Announce Type: new Abstract: Structured post-training pruning of transformers requires selecting complete functional units whose suppression causes limited degradation. We formulate structured-unit selection for language and vision transformers as a damage-aware multi-armed bandit problem under a fixed candidate-evaluation budget. Attention heads and MLP channel groups are temporarily masked on calibration batches. Paired damage is the masked loss minus the base loss on the same batch, reducing batch-to-batch variation. A smooth bounded reward drives either a UCB-style policy or fractional-Beta Thompson Sampling, and the final mask is constructed sequentially by adding one unit at each step. The selected units are functionally zeroed in the original dense checkpoint; th…

arXiv:2609.05446v1 Announce Type: new Abstract: We introduce AutoFyn, an agent harness inspired by the Expert Iteration algorithm, adapting a frozen model across many rounds by updating persistent state from verified reward signals rather than model weights. Each round begins from a fresh model session, and durable information is reintroduced only through explicit interfaces such as persistent memory files, reports, and repository state. Within a round, an orchestrator explores, plans and builds many alternative approaches with specialized agents, while a task-grounded verifier verifies the work and supplies an objective reward for measuring progress. This reward is distilled back into the persistent state, which updates the effective policy for the next round. In this technical report, w…

arXiv:2609.05441v1 Announce Type: new Abstract: Long-term memory for LLM agents is evaluated today by conversational recall benchmarks (LoCoMo, LongMemEval), which measure question answering over dialogue history, not whether remembered facts change what a tool-using agent does. We present MERIT (Memory Evaluation for Realistic Instrumented Tasks), a benchmark and harness that measures the marginal utility of memory for task-executing agents under explicit cost accounting. MERIT provides episodic tool-use tasks in three domains whose dependence on earlier-episode facts is verified by an automated leak check; a difficulty ladder ending in updated-fact recall; controlled memory corruption; and full token and dollar metering of every memory operation. Across 23,440 scored episodes ($42.57),…

arXiv:2609.05439v1 Announce Type: new Abstract: Current evaluation methods for large language models are coarse-grained and decoupled from generation, producing generic explanations that fail to provide actionable feedback for model improvement. We propose CriticGen, a fine-grained, generation-aware evaluation framework that turns evaluation into actionable control for answer improvement. CriticGen first generates sample-specific evaluation dimensions and scoring criteria under high-level categories such as subjective, objective, and self-derived constraints. These criteria then serve as a dynamic rubric for jointly producing a score, a reason, an executable refinement suggestion, and a refined answer. This rubric-conditioned refinement process enables models to diagnose flaws and perform…

arXiv:2609.05437v1 Announce Type: new Abstract: Previous AI alignment efforts have focused primarily on first-order social norms -- teaching models what is socially acceptable or unacceptable (e.g., `do not steal'). However, social intelligence depends not only on norm recognition, but also on anticipating who will enforce it and how (e.g., public shame or even imprisonment). These second-order expectations, known as metanorms, govern how people respond when social rules are broken. We introduce a novel framework for evaluating metanorm reasoning in Large Language Models (LLMs) along two dimensions: emotional appraisal and behavioral response, and propose new classification tasks, namely, predicting self-regulation in violators, and other-regulation in observers. We release a multi-perspe…

a quiet day

Discussion | Link

LandingAI has shipped Agentic Document Extraction Gen2, a rebuild of its document stack on the DPT-3 model family. Chunks are retired in favor of a document, page and block tree. DPT-3 Pro grounds to the line, DPT-3 Verity grounds to the word with a confidence score, and Parse billing now counts output characters instead of flat pages. Gen1 code will not run against Gen2 endpoints. The post LandingAI Releases Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity appeared first on MarkTechPost.

Today, we're releasing a demo of WeWorm, the first zero-click worm to spread through WeChat calls across iOS and Android. [...] The victim does not need to answer the call, or interact with their phone at all. Even if they do answer, they hear nothing, and the exploit still succeeds. [...] Working with AI, our team found the bug and wrote the first remote code execution (RCE) exploit in about two days. Building the worm took one more week. A worm at this scale used to be the kind of thing that took a larger team months. AI can already do most of the work here. Our team provided the judgment about what to target and how to test it safely. — Calif Research, WeWorm Tags: ai-security-research, ai, llms, security, generative-ai

Former Anthropic employee Jacob Coxon said AI will become ‘superhuman systems’ that can cause human extinction by the end of the decade Just a day after three Anthropic researchers warned that artificial intelligence could kill off humanity within the decade, lawmakers have begun lashing out about the risks of the burgeoning technology. Ted Cruz, a republican senator from Texas, said in an interview on ABC’s The View, that AI poses a “catastrophic risk” and that he “read that whole tweet thread that that developer put out. It was highly concerning. Continue reading...

Build and launch cloud agents with the Agents API, a managed service powered by the Codex harness for orchestration, long-running sessions, and tool use.

Together GPU Clusters now supports preemptible compute: the same GPU capacity at a flat 50% of the on-demand rate, with a five-minute drain window.

We ported ThunderKittens to NVIDIA's Vera Rubin NVL72 and rebuilt our NVFP4 GEMM around the new hardware, taking it from 42% of roofline to over 22 PFLOPS — competitive with cuBLAS and CuTe DSL. Here is what changed in the ISA and how we used it.

GPT‑Live‑1 brings natural, full-duplex voice conversations to the API, with stronger instruction following, custom voices, and telephony support.

Tool: .blend URL Viewer I'm continuing to have a lot of fun with GPT-6 Astra and Blender (see my TIL). As a big fan of the Imperial Fabergé Easter eggs, I've always thought it would be fun to make some new ones that celebrate popular culture. Yesterday I decided to try out the new ChatGPT Images 2.5 by running this prompt: Generate a photo of a faberge egg that's themed after the TV show Pluribus - research first It gave me this - honestly not bad for a first attempt! Then, just to see what would happen, I pasted that image into Codex running GPT-6 Astra (high) and prompted: Use your blender local skill to create a blender model of this faverge egg (Here's the skill file, which I created like this.) It churned away for 17m51s and built me several .blend files. I already had this vibe-code…

Google has open-sourced Mantis, a stack-agnostic toolkit of security review skills for AI coding agents. It runs the full vulnerability lifecycle: sweep the code, filter false positives, reproduce the bug in a sandbox, patch it, re-attack the patch, then score the risk. Apache 2.0, and documented as demonstration-only. The post Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities appeared first on MarkTechPost.

Learn how to deploy Qwen3.8-2.4T-A95B, a 2.4-trillion-parameter open-weight model, on Amazon SageMaker HyperPod with vLLM. This walkthrough covers cluster provisioning, NVFP4 quantization, and an OpenAI-compatible endpoint with built-in reasoning, tool calling, and native MTP speculative decoding.

Discussion | Link

Suno's new v6 AI music model is its first made with support from the record industry. Suno's Jack Brody told The Verge that v6 was "trained from the ground up, with a new set of data that does not include the same data that our previous models were trained on." The data includes content licensed from partners Warner Music Group, BMG, and Believe, as well as "user data." It's unclear whether that means v6 training data is completely free of dubiously obtained content. One of the big changes in v6 is that there are actually three different models: v6, v6-wild, and v6-mini. Mini is the model available for free to all. It's focused on fast, res … Read the full story at The Verge.

Open AI CEO Sam Altman speaks during the G20 Innovation Ministerial. | (Photo by Matt RAMEY / AFP via Getty Images) OpenAI's announcement Tuesday that it has solved one of mathematics' legendary Millennium Prize problems should have been a moment of triumph. The result is both an undeniable achievement and a striking demonstration of just how rapidly AI is transforming mathematics. But before it was even formally announced, the breakthrough had been complicated by the unusual circumstances that prompted OpenAI to pursue the problem: After hearing other researchers were making progress, it seems to have thrown its considerable resources into a last-minute effort to beat them to the punch. The ensuing controversy has surfaced allegations of scooping, spying … Read the full story at The Verg…

The companies will develop compute modules using Qualcomm Dragonwing processors, as Neura looks to scale physical AI in Europe.

At Wednesday's iPhone Duo launch event, Apple announced a handful of new Siri AI Audio Intelligence features, including Siri Recap, Live Rewind, Sound Recognition, and Music Recognition. Alongside its announcement, Apple released a document laying out how it plans to balance AI "ambient listening" and users' privacy. It says the raw audio from the new features "is handled within dedicated hardware, is not saved as a file, and is not accessible to the operating system, apps, or Apple." Apple explains that Audio Intelligence relies on the Secure Exclave in the S11 chip that powers the new Apple Watch Series 12 and Apple Watch Ultra 4: The … Read the full story at The Verge.

A weeklong summer workshop brought higher education faculty to campus to explore how AI and machine learning materials can be adapted for their classrooms.

The chipmaker is competing with Nvidia, the world’s dominant producer of AI processors.

AI models now power all manner of agents, from coding assistants that write and debug software to customer service systems The post Claude did best on a new benchmark for agents that build agents. It still passed fewer than a quarter of the tests. appeared first on The New Stack.

Discussion | Link

Discussion | Link

A recap of August 2026 launches for AI builders across Amazon Bedrock, Amazon Bedrock AgentCore, and Strands: million-token context for OpenAI models, cross-Region inference, agents that run for up to 14 days on dedicated compute, expanded AWS GovCloud availability, and Strands Robots for physical deployment.

Discussion | Link

On Tuesday evening, Anthropic pretraining researcher Jacob Coxon announced on X that he’d resigned. Within hours, two of his colleagues The post “It could kill us all”: what Anthropic’s own researchers really think about superintelligence appeared first on The New Stack.