AI News

A developer typed a TV model number into Claude Code. The agent found the set on his Wi-Fi, removed 25 Xiaomi packages, and wrote a new launcher. That is…

JR West's teleoperated ZX-2 and ZX-3 railway robots handle chainsaws and 265-pound loads at 49 feet, keeping operators safely on the ground.

A new foundation-model paper on Hugging Face claims it can cut the persistent storage needed to serve a 128K-token history by roughly 66x, by baking…

Introducing SGLang Low-Latency Mode, built with @NVIDIAAI and @AMD.

On DeepSeek-V4.1-Flash with DSpark, it delivers up to 1.38× P90 per-user token speed on NVIDIA GB300 (671 to 928 tok/s) and 1.56× on AMD MI355X (332 to 516 tok/s) in @SemiAnalysis_'s AgentX trace replay.

Highlights:
- NVIDIA: ove…

Join the discussion on this paper page

Voxtral 3: Mistral's voice agent SDK for real-time speech in 50+ languages — accurate transcription, emotional TTS, deep tool use. EU-hosted or self-deployed.

DeepInfer: High-performance, large-scale inference framework for DeepSeek models - deepseek-ai/DeepInfer

Give a face to your voice AI in 5 lines of code. Generate realtime avatar video connected to your voice AI locally - openspline/openspline

NorForge research into publicly exposed Cairn AI pentesting panels: how threat actors select targets, direct autonomous agents, exploit vulnerabilities, and expose their victims' secrets in the process.

Multimodal large language models (MLLMs) have made rapid progress on visual reasoning, yet many tasks require evidence that is difficult to obtain through internal reasoning alone: our controlled study on 22 benchmarks shows that simply increasing the inference budget of the base model yields only modest gains. Such tasks demand grounded perception, external knowledge, and numerical measurements obtained through heterogeneous tools, whose outputs must be made interpretable to the model and retained across reasoning steps. We introduce OpenVLHarness, a multimodal reasoning harness organized around a capability layer and persistent multimodal memory. The capability layer combines task-level tool interfaces with evidence rendering, providing consistent access to heterogeneous backends and converting their outputs into model-readable observations. Persistent multimodal memory retains the corresponding images, structured data, and web references as addressable artifacts throughout each query. A coding interface executes generated programs over stored data, enabling computations that combine results from multiple tools. Together, these mechanisms support an iterative process of acquiring, inspecting, and composing multimodal evidence. Evaluated on 22 benchmarks with eight open and proprietary backbones, OpenVLHarness improves average scores by 9.7-12.6 points over matched no-tool baselines, including +11.2 with Qwen3-VL-8B, +9.7 with Kimi K3, and +9.9 with GPT-6 Astra. With GPT-6 Luna at medium reasoning effort, OpenVLHarness outperforms the base model, Codex, and Visual Sketchpad at maximum effort by 7.7-8.5 points while incurring at most 37% of their model-token cost.

Agent learning is shifting from learning from human data toward learning through interaction with the world. However, scaling such learning directly in real-world computer systems is challenging and can introduce substantial security risks. We therefore propose Agent World Model in the Wild (WildAWM), a scalable pipeline for synthesizing controllable environments and tasks for training general computer-use agents (CUAs). Whereas existing approaches typically specialize in a single agent interaction interface, we construct executable environments that jointly support graphical user interfaces, model context protocol services, and command-line interfaces, allowing agents to flexibly combine visual interaction, tool use, and code execution. WildAWM is entirely powered by open-source models, demonstrating that high-quality computer-use environments can be generated without proprietary frontier models. Beyond environments, we address three recurring challenges in task synthesis: ensuring solvability, maintaining sufficient difficulty, and producing reliable verifiers, by introducing execution-based quality gates that combine heterogeneous agent rollouts, cross-judge consistency checks, and counterfactual probes of judgment robustness. The resulting pipeline synthesizes 341 long-horizon environment-task pairs that can require over 100 agent steps to solve and remain challenging even for frontier models, with GPT-6-Astra solving only 60% of them. We train two Qwen models of different sizes exclusively in these synthetic worlds. Both models achieve substantial gains across 8 out-of-distribution benchmarks spanning web, claw, and computer-use agents, demonstrating that learning in WildAWM's synthesized worlds generalizes broadly to unseen computer-use tasks. The code is publicly available at https://github.com/MSR-Orchard/WildAWM.

Banger paper from Google DeepMind on open-ended math discovery with agents.

This is one of the early, exciting works I have seen on autonomous AI-powered discoveries.

(bookmark it)

A Gemini-based agent found and formally verified new acceleration rules for eleven unresolved Turing machines in BB…

Contribute to google-pytorch/torch_tpu development by creating an account on GitHub.

Eka Robotics may be the first robotic arm that is actually good at picking up a wide range of objects, and it is so smooth and human-like, it is wowing the industry.

🚀 Full-Duplex Interaction with SGLang-Omni

We've expanded SGLang-Omni's support for real-time interaction with a unified runtime abstraction for:
- Native full-duplex models
- Turn-based multimodal models
- Cascaded systems

In single-session tests on 1× H200, the Qwen3.8-27B cascade (thin…

Large language models (LLMs) can influence user trust and decision-making through subtle linguistic cues, yet the behavioral effects of manipulative outputs remain poorly quantified. We introduce a behavioral computing framework for measuring \textit{behavioral vulnerability}, defined as users' susceptibility to manipulative AI responses. Building on DarkPatterns-LLM benchmark, we generate controlled manipulative stimuli using Manipulation Gradient Descent (MGD), targeting biases associated with authority, urgency, and social proof. We analyze participant responses using the Multi-Stakeholder Impact Assessment Network (MSIAN) to model cross-user influence and the Temporal Harm Propagation Model (THPM) to characterize the persistence of behavioral effects over time. In a study involving 120 participants, manipulative framing produced significant changes in trust calibration and compliance ($p < 0.01$), with effects also propagating across social contexts. These findings highlight the need to extend LLM safety evaluation beyond model outputs to measurable behavioral consequences and provide a computational framework for quantifying individual, social, and temporal risks arising from manipulative AI interactions.

Large language models (LLMs) can leverage semantic feature names alongside summary statistics for tabular feature selection, yet it remains unclear when such information improves subset selection relative to established statistical methods. We investigate this question within an 18-dataset framework comprising 10 datasets for Group Relative Policy Optimization (GRPO) training, 2 for checkpoint validation, and 6 held-out evaluation datasets. We compare five classical baselines, frozen Qwen2.5-7B and Llama-3.3-70B selectors, and two GRPO-adapted Qwen2.5-7B models, reporting classification ROC-AUC and regression R^2 separately and using dataset-wise ranks for cross-task comparison. Across five seeded evaluation partitions, classical selectors generally achieve higher predictive performance than the tested 7B LLM/RL selectors, including under a pre-specified ~33% feature budget with strict train-only classical fitting, while the 7B selectors retain compact subsets. A cardinality-matched semantic sensitivity experiment shows that, for Zero-shot Qwen-7B, informative feature names improve regression performance over anonymized and shuffled alternatives (+0.1964 and +0.1606 mean R^2), whereas aggregate classification differences are smaller (+0.0159 and +0.0084 ROC-AUC) and dataset-dependent. These results characterize when semantic information benefits LLM-based feature selection and show that classical selectors remain strong baselines for predictive tabular feature selection.

We’ve acquired Orchestra Bio, a platform that helps biopharma teams coordinate the complex process of drug development.

Together, we’ll build on Rosalind, our AI models and tools for life sciences research, to streamline those operations and help new treatments reach patients sooner.

Welcome @ash…

We’re releasing all 211 of the tested designs and the lab data as a resource for scientists: https://t.co/zqke30UV2o

Read the full technical report: https://t.co/ikWVcFxnyP

Mac.apk: Run Android apps natively on Apple Silicon Macs - kksimp/Mac.apk

Frontier agents retrieve structural metadata well and compute geometry from coordinates less reliably. We ask how much of a benchmark score reflects that gap and how much reflects the evaluation configuration around the model. We study ProteinGeoBench, a suite of 131 protein-structure tasks, using 4,249 rollouts from four frontier models and 4,486 further trials across 26 models. Three of the four frontier models fall within 0.031 of one another (0.785 to 0.816) with no difference between them resolved, while the fourth scores 0.707 and all three gaps to it exclude zero. Three of the four do better on metadata lookup than on numeric calculation. The fourth is slightly better on numeric. Numeric tasks are 28 of the 131 but half of the twelve tasks that all four models fail. The ranking also depends on the evaluation budget: with one run per task, 23.9% of bootstrap replicates put a different model first. On a selected 15-task slice, removing network access lowers scores more than blocking eight structural-database domains, but the two contrasts were graded in different passes and cannot be cleanly decomposed. Swapping in each model's own command-line agent changes scores by -0.071 to +0.086, a comparison too small to rule out moderate harness effects. Across 26 models, numeric performance rises with non-numeric performance rather than staying flat. The experiments use different task subsets, so they cannot rank model, harness, access and task composition against one another. Evaluations should report results by task family, along with the number of runs, the harness and the access conditions.

Safety alignment for large language models (LLMs) is primarily performed during post-training, while recent work has also begun incorporating safety data into pre-training. Yet safety can degrade when later training no longer includes safety data, and gains in safety often come at the cost of general capability. This tension suggests that safety and general reasoning should not be treated as separate training goals. We instead view safety as an LLM capability that can be learned through a reasoning structure shared with general tasks. Both require the model to understand intent or task goals, apply relevant knowledge, and reason about potential consequences. We introduce safety mid-training, a stage between pre-training and post-training that jointly develops safety and general reasoning through this shared structure. Given safety and utility prompts, the base model generates structured documents and reasoning traces that are then used for mid-training. Safety mid-training reduces average attack success rate (ASR) by 72% to 81% relative to Qwen3-1.7B, 4B, and 8B base models, while also improving general capability. The benefits persist through subsequent post-training. After general-only SFT, mid-trained models largely retain their safety, achieving lower ASR than the baselines without sacrificing general capability. They also retain safety under general-only GRPO without safety rewards, while subsequent RL reaches a better safety-utility frontier when initialized from the mid-trained models. Together, these results suggest that learning safety and general reasoning within a shared mid-training stage can reduce the safety-utility trade-off and make safety more robust to subsequent training.

Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2$\times$ less pretraining compute. Further, Halcyon-1.6B outperforms JiT-T2I-XXL/8++ (35.3 vs. 41.0 gFD) with 275$\times$ less pretraining compute. Controlled experiments show that text-only pretraining does not improve T2I generation. Halcyon also learns image captioning by replacing caption dropout with I2T supervision, maintaining competitive T2I generation quality under the same pretraining budget. We show that strong T2I systems can be trained from scratch without internet-scale language pretraining. Code and models are available at https://cs.stanford.edu/~keshik/halcyon