New

Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Controlarxiv.org

Zing‑0.5 is a 5 B autoregressive world model that lets users control generated environments in real time using both keyboard actions and text prompts. The paper introduces unified action‑text conditioning, segment‑level teacher distillation, and a low‑cost streaming inference pipeline that runs at 24 FPS (832×480) for about $0.009 per minute, achieving 81 % overall and 88.5 % consistency on a nav…

What is CIAM in 2026 and why does it matter?twilio.com

CIAM in 2026 has moved from static login checks to a continuous trust layer that must handle AI agents, deep‑fake attacks, and zero‑friction authentication, requiring a split between an identity authority and a real‑time trust orchestration layer.

How to Orchestrate Multi-Call Conversations with an LLM and Twilio Conversation Memorytwilio.com

A step‑by‑step tutorial for building a C# ASP.NET Core service that uses Twilio Conversation Memory and OpenAI’s gpt‑4o‑mini to retain caller context across multiple phone calls, with full code snippets, environment setup, and prompts for both real‑time response streaming and call summarization.

Native Drag and Drop Meets Cross-Device Continuitycodenameone.com

Codename One 8.0 adds two user‑experience primitives: **Cross‑Device Continuity** (state checkpointing + Apple Handoff + custom HTTP relay) and **Native Drag‑and‑Drop** (OS‑level drag using the same ClipboardContent model). The post shows the API (StateProvider, Continuity.checkpoint, NativeDragOperation), platform support tables, and practical advice on payload size, security, thread handling, a…

How to operate shared platforms safely at agent scaledatadoghq.com

Datadog explains how scaling AI agents turns isolated executions into shared‑platform risk and outlines a systematic approach to model agent trajectories, monitor per‑dependency constraints, and enforce workload‑specific capacity policies. The result is proactive detection of bottlenecks and protection against noisy‑neighbor failures.

FLAT: Resampling Image and Text into 1D Flexible-Length Aligned Transmodal Tokens for Retrieval and Generationarxiv.org

FLAT is a pre‑training framework that encodes images and text into a shared 1D token sequence with variable length via nested dropout, enabling the same embeddings for cross‑modal retrieval and generation. It attains state‑of‑the‑art scores on COCO and Flickr30K for captioning, retrieval, and text‑to‑image generation, and supports interpolation and zero‑shot composed retrieval.

The only perfect Endpoint Prevention and Response (EPR) score in 2026 belongs to Elasticelastic.co

Elastic Security achieved a perfect 100% score in AV‑Comparatives’ 2026 Endpoint Prevention and Response (EPR) test, stopping all 50 attack scenarios at the prevention stage with zero false positives or workflow delays. The test measured active prevention, passive detection, operational accuracy, and workflow impact across 14 vendors. Elastic’s claim is backed by open‑source protection rules and…

Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agentsarxiv.org

EvoSkill‑GUI lets GUI agents revise their procedural skills on‑the‑fly without extra training by using a reflect‑revise‑reuse loop that edits skill packages during execution. The approach yields up to +16.2% improvement on MobileWorld and similar gains on AndroidWorld and OSWorld, and the evolved skills transfer to related tasks.

Your Donations at Work: One Year of Sponsored Servo Developmentservo.org

A year‑long, donation‑funded effort by Josh Bowman‑Matthews on the Servo project resulted in 8 new maintainers, 1150 PR reviews, 114 contributor‑focused issues (92 % resolved), extensive new documentation, and concrete stability work on the JS engine integration and flaky tests, plus support for a community grant.

RiskChainBench: A Benchmark for Obfuscated Platform Message Restoration and Evidence-Grounded Web Investigationarxiv.org

RiskChainBench is a new benchmark that pairs synthetic obfuscated message restoration inputs with human‑labeled local web environments, requiring models to both decode malicious instructions and investigate the linked site. Across ten models, restoration accuracy varies widely and web‑agent failures dominate the error budget.

EventEgoHands++: Event-based Egocentric 3D Hand Mesh Reconstruction with Real Datasetarxiv.org

EventEgoHands++ adds an instance‑aware hand detector and an Adaptive Attention module to event‑camera egocentric hand‑mesh reconstruction, distinguishing left/right hands and modeling their interaction. The authors also release EEH‑R, a ~1 M‑frame real dataset, and report consistent gains over prior baselines on synthetic and real tests.

Gemini 3.8 Live models now available on AI Gatewayvercel.com

Vercel AI Gateway now offers Google’s Gemini 3.8 Live and Live Extended Thinking models through its realtime WebSocket API, enabling real‑time spoken interactions with audio output and parallel reasoning. The post includes a short TypeScript example for token acquisition, WebSocket setup, and handling of realtime events.

Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agentsarxiv.org

XConf (eXperiential Confidence) is a novel method for estimating language model confidence by leveraging the model's accumulated experience from past graded episodes. It significantly outperforms existing methods like self-consistency in discrimination and calibration, at a fraction of the computational cost, across various tasks.

PhysStream: Streaming Physics-Grounded Video Generation with Structured Scene Memory and Fine-Grained Motion Controlarxiv.org

PhysStream introduces a two‑stage autoregressive video generator that uses online‑derived positional and tracking maps (structured scene memory) and sparse velocity‑increment signals to enable fine‑grained, physics‑grounded control of multi‑object tabletop scenes. It cuts motion distribution error by 33 % and trajectory error by 12 % versus strong baselines, and wins 85 % of human preference test…

Fingers as Legs: Learning Self-Supported Locomotion and Manipulation with an Anthropomorphic Handarxiv.org

The authors train an anthropomorphic robotic hand to crawl, steer, and recover from falls using its fingers for both support and manipulation, via a reinforcement‑learning reward formulation tuned to the hand's asymmetry. Sim‑to‑real experiments show faster locomotion than quadruped‑style rewards and successful untethered tasks without onboard vision.

Monitor TAS and gang scheduling for AI training in Kubernetesdatadoghq.com

Kubernetes’ default scheduler can’t satisfy AI training’s need for low‑latency GPU interconnects and simultaneous pod start‑up. The blog explains how the open‑source Kueue job queue adds topology‑aware placement (using node labels like `topology.kubernetes.io/rack`) and how the Coscheduling plugin adds a permit phase that only binds a gang of pods when the full set is ready, preventing idle GPU r…

Verifiable Social Reasoning for LLM Assistantsarxiv.org

The paper introduces Fuse, a multi‑agent simulation that gives LLM assistants a verifiable ground‑truth task for social reasoning by hiding a target agent’s motive and letting a user‑mediated conversation infer it. Experiments on 12 LLMs show user mediation makes reasoning harder, models are biased by user framing, need more detail than humans, and longer chats don’t always help.

ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signalsarxiv.org

ImpossibleRubrics introduces a 169‑task benchmark of “impossible” prompts, each with a formal oracle certificate defining what an honest answer can claim. The authors generate rubrics downstream and test them adversarially, finding that many rubric generators are exploitable (8‑36% of the time) and that a single generic rubric (“be decisive, penalize hedging”) is exploited 64% of the time, while…

ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agentsarxiv.org

ScienceBuddy is an interactive workspace that converts researcher prompts, feedback, and execution traces into continual‑learning tasks for AI agents. It introduces a "recursive‑in‑recursive" self‑improvement loop that alternates harness refinement and model training, and showcases case studies across four scientific task families.

9 top conversational AI platforms in 2026twilio.com

Twilio’s blog lists the nine leading conversational‑AI platforms for 2026, highlighting how the market consolidated and what capabilities matter when choosing a vendor. It details Twilio’s own infrastructure tools—ConversationRelay, Agent Connect, Orchestrator, and Memory—showing sub‑second latency and full‑stack reliability for voice and messaging.

Rails Hyperdrive: supercharged agentic development for Railsevilmartians.com

Rails Hyperdrive is a development‑only Rails engine that mounts a Model Context Protocol (MCP) server and lets any gem ship agent knowledge (skills, agents, commands, guidelines) as markdown files. The engine adds no knowledge itself; companion gems provide the context‑aware artifacts that agents can query at runtime, avoiding token‑heavy guessing and stale schema reads. Installation is a single…

Manage Cursor costs with Datadog Cloud Cost Managementdatadoghq.com

Datadog Cloud Cost Management now integrates Cursor AI‑coding usage, exposing per‑user, per‑model, and mode breakdowns, out‑of‑the‑box dashboards, anomaly detection, and budget/monitoring tools so FinOps can track and control AI coding spend alongside other cloud and SaaS costs.

AI for Games in the Foundation Model Eraarxiv.org

The paper surveys how foundation models are used across six roles in the game development lifecycle—from playing agents to design assistance and runtime adaptation. It highlights limited transferability due to game-specific interfaces and notes that evaluation is mature for bounded play but weak for adaptive and testing scenarios.

TAPe+ML: A Compact Structured Representation for Multi-Task Computer Visionarxiv.org

TAPe+ML v3 is a compact multi‑task vision system that replaces raw‑pixel processing with a structured TAPe representation. Using <100 k parameters, it achieves 84.7 mAP50 (65.3 mAP50‑95) on COCO detection, 80.7 mask mAP50 (58.4 mask mAP50‑95) on COCO segmentation, 92 % top‑1 on Imagenette and 89.9 % on ImageNet‑Real, while also showing robustness to distribution shift in video scene detection.

LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligencearxiv.org

LimiX-2 is a new tabular foundation model that replaces the usual target‑centric in‑context learning objective with a joint‑distribution objective via Contextual Mechanism Networks (CMNs). Trained on synthetic causal graphs using Context‑Conditional Masked Modeling, it outperforms prior tabular PFNs on TabArena, TALENT, and BCCO and can recover causal skeletons from attention patterns.

Kythe, a pluggable, language-agnostic ecosystem for building tools that work with codekythe.io

Migrating to Java 17: The Hows, Whys, and Whenscloudbees.com

A product‑focused overview of CloudBees CD/RO’s new integration with Argo Rollouts, describing the supported blue‑green and canary strategies, service‑mesh compatibility, and UI/analytics features. No deep technical walkthrough or performance data.

Running Ubuntu on the Lenovo IdeaPad Duetvhaudiquet.fr

The author swaps postmarketOS for an Ubuntu rootfs on a Lenovo IdeaPad Duet, discovers the device boots via Depthcharge using a FIT image, and explains why a UEFI‑capable firmware like U‑Boot is needed to run Ubuntu's kernel.

Show HN: Pull every comment out of a Google Sheet, in the browserbensunter.com

Turn Arduino® UNO™ Q into your local 3D printing watchdogarduino.cc

A guide details how to build a local 3D printing watchdog using an Arduino UNO Q and a camera. It leverages Edge Impulse's FOMO-AD model for anomaly detection, pausing prints via Moonraker API without cloud connectivity.