AI News

Banger paper from Tencent on recursive self-improvement for terminal agents.

(bookmark it)

This is a great read if you are tracking research on training terminal or coding agents with RL

Starting from Qwen3.8-27B, they alternate two phases over five stages. One phase fine-tunes on rewritten vers…

Il 56% delle PMI italiane investe in AI meno dello 0,5% del fatturato e solo il 4% arriva a trasformare il proprio modello di business

I controlli dell’AI Act in Europa coinvolgono circa duemila autorità. Ecco chi vigila, quali sanzioni scattano e il modello italiano

Pin MCP tool contracts, catch silent drift. Field audit of the npm MCP top: 218 silent contract changes (Microsoft, Google, Hostinger included) - receipts and weekly CI watch inside - rugsnare/docs/blog/2026-10-17-mcp-contract-drift.md at main · Paraphern/rugsnare

a diff viewer for the age of aquarius. Contribute to andunieee/twain development by creating an account on GitHub.

Pick hardware and a model, and feel the real speed of a local LLM before you buy. Simulates tokens/sec and time-to-first-token from verified benchmarks for GPUs, Macs, and more.

Unified multimodal generation covers discrete, continuous, and hybrid paradigms. Discrete approaches with single-codebook visual tokenizers align naturally with text modeling but have limited representational capacity. Continuous and hybrid approaches retain richer visual representations, but they model text and images in heterogeneous spaces with different training objectives, potentially introducing optimization conflicts between modalities with misaligned representation spaces. These choices expose a central question: how to efficiently model high-capacity visual representations alongside textual content in a shared optimization space within a single backbone. We present Intern Lumina U2, to our knowledge the first diffusion language model to represent multi-codebook visual tokens in a fully discrete space. Each visual position is represented by a group of codes from complementary codebooks, increasing its representational capacity while preserving compatibility with discrete text modeling. At the input, the embeddings are concatenated at each position; at the output, a lightweight multi-codebook autoregressive head predicts the codes within each position sequentially, while the diffusion backbone processes all positions in parallel. Built on the unified 4D representation of AToken, Intern Lumina U2 supports image, video, and 3D inputs within a single backbone and achieves leading performance among unified models in visual understanding, image generation, and image editing. We further build a full-stack training, inference, and evaluation system on both Ascend and NVIDIA platforms to verify the cross-platform effectiveness of our model. We also improve the inference efficiency through a stability-aware sampling strategy, which combines confidence-plateau-guided unmasking approach with drift-triggered cache refreshing mechanisms, accelerating image generation by 4.45x.

Online reinforcement learning (RL) with verifiable rewards has driven rapid progress in mathematics, coding, and tool use, but scaling this success to GUI agents remains challenging due to the high cost of interactive environments and the scarcity of realistic training tasks. We introduce MAI-Gym, an open all-in-one platform that unifies GUI agent evaluation, data collection, and online RL training across mobile, computer, and web environments. Through a shared rollout pipeline, MAI-Gym ensures consistent agent behavior across evaluation and training, supports ten established benchmarks, and scales to 20,000 parallel environment instances. Beyond scalable infrastructure, MAI-Gym enables automated synthesis of diverse GUI applications and challenging tasks, including resettable mocked apps and coherent cross-app worlds with execution-validated task verifiers. Using MAI-Gym, we train MAI-Gym-9B, a unified 9B agent for mobile and computer use, achieving state-of-the-art performance among open GUI agents under 10B parameters: 78.4% on MobileWorld and 76.2% on OSWorld-Verified. Importantly, RL in a mocked train-ticketing app transfers to its real-world counterpart, improving task success on a phys- ical phone from 28.8% to 48.1%. We release the framework, synthesis skills, mocked apps, 2,000 verified tasks, and the model checkpoint to support repro- ducible research and further advances in scalable GUI agent training.

Venture capital continues to pour into fintech, AI infrastructure, semiconductor manufacturing, healthcare, and physical AI over the past 24 hours, with Ramp leading global funding activity after raising $1.85 billion at a $60 billion valuation. The New York fintech company accounted for most of the disclosed dollar-denominated investment, but the day's other significant financings reveal

Taiwan considers stricter export controls on AI chips to China, aligning with US efforts amid Beijing's rapid AI expansion.

Studio IBM: AI agentica, il 78% dei dirigenti italiani segnala incidenti AI ai vertici, ma solo il 37% rivede la governance. Costi e norme

jq, but for AI streaming: read OpenAI-compatible chat completion streams at a glance, with TTFT and timing - vivgrid/sjq

Navige uses Sufficiency Intelligence to cut AI agent costs while work runs. It sends suitable requests to smaller models, caches repeated reads and trims bulky tool results. If unsure, it keeps your chosen model. Set budgets, catch runaway loops and approve sensitive actions. See costs and savings by job, with estimates clearly labelled. Watchtower is free forever. Preview changes before switching controls on.

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Polimi: cloud italiano a 9,48 miliardi nel 2026 (+17%): l'AI accelera il mercato, ma la sovranità europea pesa sulle scelte delle imprese.

We're pleased to announce that the Advantage2™ system is now generally available via the Leap™ quantum cloud service and for on-premises installation. This annealing quantum computer is a powerful and energy-efficient system capable of solving computationally complex problems in business and science. Designed to tackle highly complex computational problems and facilitate quantum and hybrid-quantum applications for production deployment, the Advantage2 system’s key benefits include: • Increased qubit connectivity: The Advantage2 processor features the Zephyr™ topology with 20-way connectivity, which enables embedding of more complex problems. • Higher energy scale and lower noise: The Advantage2 system offers a 40% increase in energy scale and a 75% reduction in noise, which contributes to higher-quality solutions for complex calculations. • Greater coherence: A twofold increase in coherence allows the Advantage2 quantum processor to achieve faster time-to-solution. • Energy-efficient processing power: D-Wave’s quantum computers, including the Advantage2 system, have required the same amount of electricity over six generations—a mere 12.5 kilowatts. • Fast anneal: This feature enables coherent quantum annealing at scale, greatly reducing the impact of external disturbances such as thermal fluctuations that can hinder quantum calculations. • Hybrid solver integration: Now integrated with the full Advantage2 system, the hybrid solvers in the Leap quantum cloud service—including the powerful nonlinear hybrid solver—support up to two million variables and constraints, allowing businesses to run large-scale, business-critical applications in production. The Advantage2 system is immediately available for customer use via D-Wave’s LeapTM real-time quantum cloud service and on-premises installation. To learn more about the Advantage2 system and start using it today, visit: https://www.dwavequantum.com/solutions-and-products/systems/ #QuantumRealized #QuantumComputing #DWave #Advantage2 #AdvantageSystem #Innovation #Technology #Launch #Announcement #Optimization #AI #MaterialScience #Milestone

Multi-agent context firewall, MCP schema tree-shaker, and token budget guard for Claude Code (CLAUDE.md), OpenAI Codex (AGENTS.md), and Cursor. - FramY2/ctxwise

Few-step autoregressive video generators enable streaming generation, yet they run every token through the final denoising step, regardless of how much refinement it needs. We find this uniform computation is often unnecessary: many tokens already closely match their eventual full-step outputs in the initial steps, while others require refinement until the end. We introduce Speculative Forcing, a framework that treats intermediate clean estimates as drafts and predicts when individual tokens can be safely accepted early. Once accepted, a token is frozen as its final output and undergoes no further denoising, while its cached keys and values remain available as context for the remaining tokens. We explore two alternative verifier designs: a training-free one based on similarity to clean content and stability across steps, and a learned one that reads the frozen generator's intermediate features to estimate each token's remaining gap to its full-step output. Across short-video, long-video, and world generation, Speculative Forcing preserves the quality of full-step sampling while substantially reducing computation. On LongLive, it matches the four-step generator with ~1.7 denoising steps per token, running over 1.5x faster at 41.7 and 37.8 fps. On Lingbot-World 2.0, where camera motion continually changes the visible content, it maintains quality with ~2.5 denoising steps per token and runs 1.35x faster at 24.5 fps.

Computer use has reached a turning point: the focus is expanding from GUI-only interaction to systems that complete tasks on users' computers through visual and programmatic interfaces. We define a generalist computer-use agent (GCUA) as a system in which the model chooses how to observe, act, and organize its work across visual and programmatic interfaces. Yet frontier implementations, such as GPT-6 Astra's computer-use system, remain largely closed, limiting systematic study of their design. We introduce Open Generalist CUA (OpenGCUA), an ongoing project to develop and open the full stack for GCUAs with the community. This report presents the first release of OpenGCUA. Code execution is a key feature of our agent design: it lets intermediate results determine subsequent operations and enables models to apply their existing programming capabilities directly to computer interaction. We also propose harness-to-model decision transfer, moving selected observations and task-state decisions from fixed harness rules to the model. On OSWorld-v2, our full configuration improves task scores across all eight evaluated models, including GPT-6 Astra and Claude Opus 5, raising mean partial score by +19.6 points and reducing reported steps per task by an average of 53.0% relative to batched GUI-only execution. We release our implementation and prompts, evaluation trajectories from all eight models, and preliminary versions of a real-device harness and a cross-device SDK. The harness addresses real-device challenges not fully captured by benchmarks, while the SDK provides a common programming interface across computers, mobiles and browsers. Across four real macOS tasks under our real-device harness, selected Qwen3.8-Max runs use 19.4 times as many reported reasoning tokens as GPT-6 Astra, highlighting substantial differences between models in real-device execution efficiency.

Most existing agentic reinforcement learning methods assume single-session trajectories, where interaction proceeds within one continuously accumulated context. However, agentic techniques such as context compaction and subagent forking may produce trajectories spanning multiple sessions, creating a mismatch between the single-session training assumption and inference-time execution. Therefore, we introduce Multi-Session Agentic Reinforcement Learning, a more general setting that supports an arbitrary number of sessions within a trajectory. Under this setting, a widely adopted method is to broadcast trajectory-level advantages to all sessions, but this does not distinguish contributions from various sessions or guide session allocation within trajectories. To address this issue, we propose Mosaic, a multi-session agentic RL method with critic-free hierarchical optimization. At the trajectory level, it incorporates session-usage signals into advantage estimation. At the session level, it shapes each session's advantage according to the estimated marginal contribution. Experiments on BrowseComp with Qwen3.5-4B show that Mosaic outperforms broadcasting in both context compaction and subagent forking scenarios while using fewer sessions on average. Further analyses show that our method achieves improvements in session utility and allocation, which we suggested is a promising target for future advances in multi-session agentic RL.

America’s next top model, Reflection’s Beam, is now live on Together AI.

Trained on Together GPU Clusters, Beam is Reflection’s first open-weight model.

• ~501B-parameter MoE, 23B active
• Built for coding + long-horizon agentic work
• 1M-token context window
• Released under an Apache 2.0 releas…

Si profila un biforcazione strategica tra superintelligenza sovrana gestita dallo Stato e modelli proporzionati e sostenibili per il mercato

The Higher Order Harness (HOH "/oʊ/" ), projected as a unified chat interface. - longsurf-ai/endo

Official Z.ai platform to experience our new, MIT-licensed GLM models (Base, Reasoning, Rumination). Simple UI focuses on model interaction. Free.