AI News

ICYMI: Lightx2v just dropped Qwen Image 2.1 Lightning ⚡️

> 4-step distilled LoRA
> weights: https://t.co/QPXZgrfwvL
> space: https://t.co/K9Sx7MZyw5

Join the discussion on this paper page

Training language agents is bottlenecked by the high cost of long-horizon rollouts, yet standard agentic reinforcement learning (RL) learns only from a terminal scalar reward, leaving the supervision in intermediate steps largely unused. Prior work exploits this supervision through value targets that estimate per-step success probabilities or observation targets that fit environment response. Both are hard to construct well: (1) value targets come from the same single scalar signal for each step in the trajectory, and (2) observation targets, such as web pages and terminal logs, are long and noisy and reach only one step ahead. We propose \name, which adds a third kind of multi-step action targets. On successful on-policy trajectories, \name trains the policy backbone to generate the action sequences that follows each history. Each action is also supervised when forecast from the preceding histories, forming a temporal ensemble that the shared backbone absorbs during training, and this enables acting and forecasting to co-evolve. The policy still acts one step at a time at inference, with no overhead. Across five benchmarks (ALFWorld, ScienceWorld, WebShop, $τ^2$-bench, and AppWorld), \name (1) achieves stronger aggregate performance than GRPO and RL baselines with predictive auxiliary targets across simulated environments, while improving overall performance on both realistic applications; (2) delivers up to $1.9\times$ and $2.0\times$ higher sample efficiency than ECHO and GRPO, respectively, on ScienceWorld; and (3) achieves these gains with a single policy backbone, without additional models, separate pretraining, or inference overhead. The codebase is available at https://github.com/ulab-uiuc/CoPE.

At this week’s OCP Global Summit, new details emerged on Intel’s upcoming data center GPU and a new kind of AMD EPYC CPU while Nvidia made big open-source software moves.

OpenCode 2.0.

The problem with giving everyone in a company access to AI tools is that people will use them in unexpected ways. And that fact is steadily seeding fears that companies now have new, gaping security holes in their systems or, at least, will soon.

MCP server and CLI that crawls websites for AI agents: finds the answer page 2x more often than Crawl4AI, 10x cheaper per answer than letting Claude crawl. - RayenMlayeh/jevcrawl

A budgeting app with AI on your own machine. Your ledger and your questions both stay on disk. The models download once, then nothing about you leaves.

Promise Robotics has signed a multi-year partnership with America's largest homebuilder by volume to bring physical AI to construction.

Learn to build custom harnesses, folks!

Here is an impressive paper showing the impact of a good custom harness on long-context work.

GLM-5.3-Flash goes from 4.4% to 61.1% exact match on OfficeQA Pro V2 with the same model.

The authors report that the gain comes from the harness.

The model-agno…

How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge these two directions with Hybrid Reasoning Agents (HybridAgents), which dynamically navigate between latent reasoning and text-based interaction throughout a unified generation process. Architecturally, a lightweight HybridLink enables seamless transitions between text and latent modes, while the agent's continuous working memory preserves lossless hybrid context across multiple turns. These designs preserve tool-integrated agentic capabilities while making latent thoughts directly inspectable. Building on this architecture, we further develop dedicated post-training algorithms for hybrid reasoning: HybridSFT stabilizes tool-integrated hybrid generation by teaching the backbone to coordinate with HybridLink, then HybridRL scales reasoning by learning when to think in latent space and when to act through text. We evaluate HybridAgent at 3 model scales across 11 benchmarks covering mathematics, science, medicine, search, and agentic reasoning. Compared with frontier LLM Agents and agentic frameworks spanning both text-based and latent methods, HybridAgent consistently improves performance by 10.4%, while delivering 35.5%-54.7% token reduction, 1.8x-2.6x faster inference, and 2.7x-3.7x cost savings. Code and data are available at https://github.com/togethercomputer/HybridAgent.

Computer-use agents have demonstrated strong capabilities in interpreting screens and executing keyboard and mouse actions. However, games, with their visually rich worlds and diverse mechanics, remain challenging and offer a demanding next step toward general digital control. Current game-agent learning pipelines often rely on large-scale demonstrations with limited coverage of dynamic game states, while scalable environments for interactive training remain difficult to obtain. We introduce Player-1, a post-training framework for learning game control beyond demonstrations through synthetic experience. We extend the computer-use action space with relative pointer motion, sustained inputs, and composite actions. For supervised initialization, we fine-tune computer-use models on 6,044 curated gameplay trajectories that pair recorded actions with task instructions and visual reasoning. We construct Player-1-Gym, sixteen synthetic 2D and 3D environments designed around control demands shared across games. Programmatically verifiable rewards and randomized observations and control dynamics support multi-environment reinforcement learning with PPO. The resulting Player-1-9B and Player-1-35B-A3B improve MineDojo-300 success over their SFT policies by 27.8 and 17.7 percentage points, respectively. Player-1-9B reaches 38.33% success compared with 12.63% for the strongest evaluated baseline, with further gains over SFT on ViZDoom-10 and GameWorld. Further analyses show that broader synthetic environment coverage improves transfer, while the use of composite actions that combine multiple keyboard and mouse inputs increases during RL. We demonstrate that learning from synthetic experience substantially improves game control beyond supervised initialization, with gains that generalize beyond the RL training suite.

Join the discussion on this paper page

Mid-2026 reporting on autonomous AI evaluation incidents claimed that frontier large language model (LLM) agents escaped server containment through emergent coordination and cloud infrastructure attacks. This paper contextualizes those claims through forensic analysis and containerized recreation of both the Highly Persistent Internal Model (HPIM) training-case conditions and the Phase One privilege escalation targeting internal package repositories during OpenAI's ExploitGym evaluations. We recreate the configuration of the OpenAI-provided training example and reproduce the cryptographic verification bypass (CVE-2026-65616) in isolated repository servers, where misconfigured systems enabled token forgery and privilege escalation through basic HTTP requests, Python, and JavaScript. Crucially, the training-example conditions and the Phase One (and later) misconfigurations share the same exploitable patterns. Our analysis explains how Zero Trust Architecture principles neutralize this shared attack vector. Contemporary AI risks derive directly from training and deployment conditions in ambivalent learning environments lacking adequate containment measures.

New impressive work from Microsoft and collaborators.

And it's one of my favorite recent topics: RL environments.

They show that Qwen3.6-35B-A3B goes from 41.1 to 53.3 on Terminal-Bench 2.1 by training on environments it designed itself.

AutoEnvScaling gives environment design to a proposer agen…

New York-based healthcare AI startup Brellium has raised $23 million in venture funding to help healthcare providers catch medical billing errors, identify gaps in patient care, and avoid costly insurance audits before problems become financial losses. The company announced the funding alongside the launch of CortexLM, an AI system that reviews patient records for documentation errors and

Out-of-process authorization proxy and safety gateway for autonomous AI agent tool execution. - mohitdebian/AgentGate

An open-source benchmark of how much code AI models add.

Max Hodak, the former president and co-founder of Neuralink, detailed onstage at TechCrunch Disrupt how his company's work with PRIMA is part of its path toward a novel brain-computer interface (BCI).

Gemini 4 Argon taught a robot to play cornhole just by watching a video.

It mapped the physics, coded its own simulation, and taught @Apptronik's Apollo humanoid to sink 7 throws in a row.

This is my new favorite way to keep up with AI papers.

I turn my curated papers into a 2-minute spoken technical briefing with @FishAudio Drama 3, straight from Codex.

In the demo, I took one of my favorite recent NVIDIA papers and built a small skill that reads the full paper, writes the script,…