Live intelligence

Latest AI updates

6 updates
Research

ZoomGain builds fine-grained visual supervision from global-to-zoom perception gaps

ZoomGain is a data construction pipeline that creates questions about a full image using visual evidence observable only after zooming in. It verifies answers on the zoomed view with multiple models, checks that questions remain unambiguous in the full image and cannot be answered from language priors, and generates multiple-choice distractors. The authors build ZoomGainBench, a 1,000-question bilingual benchmark where the best of 28 tested VLMs scores 61.08, and report that training on 9K samples improves 4B/9B models, with the 9B model reaching 90.25 on HR-Bench 4K.

Vision-Language ModelsFine-Grained Visual PerceptionBenchmark Construction
Research Editor selected

Shuffled OPSD: When Semantic Consistency Hurts On-Policy Self-Distillation?

This paper introduces Shuffled OPSD, a variant of on-policy self-distillation that shuffles the words of the shared problem statement while keeping the teacher-only worked example intact. Across three math reasoning benchmarks and three Qwen3 model scales, it improves mean accuracy over both intact-target and Standard OPSD, with the largest gain of 3.52 points on Qwen3-1.7B. Controls show that shuffling the teacher-only example hurts accuracy, while word-level shuffling of the problem statement helps, challenging the assumption that semantic coherence is always beneficial.

Why it matters: Provides concrete, recent experimental evidence that disrupting input structure can improve self-distillation, offering a novel insight for AI training methodology.

on-policy self-distillationknowledge distillationmathematical reasoning
Research

CoworkOS: A Harness for Long-Horizon Agent Coworkers in Real-World Automation

CoworkOS is a general-purpose harness for long-horizon work that separates task strategy from execution management, enabling agents to handle complex productivity tasks with human-in-the-loop participation. It features a three-layer architecture, hierarchical workspaces, and persistent semantic memory for continuity across sessions. In benchmarks, it is competitive with frontier agent applications, and a case study shows automated post-training improving model scores by +9.76 over five rounds.

AI agentsautomationhuman-AI collaboration
Models Editor selected

Axion 1.0: Open Agent Foundation Model with Simplified Training Recipe

Axion 1.0 is a fully open agent foundation model trained in just three stages: two supervised fine-tuning stages and reinforcement learning, avoiding complex techniques like DPO or on-policy distillation. It achieves competitive performance on agentic benchmarks and releases the complete training recipe, including data construction and code, for full reproducibility. The model demonstrates that strong agentic capabilities can be built with a simple, open approach.

Why it matters: Provides a concrete, recent AI development with a novel simplified training method and full open-source release, valuable for practitioners and researchers.

agentic AImodel trainingopen-source AI
Products

AWS announces SOCI-indexed images for faster SageMaker AI job startups

AWS has announced Seekable OCI (SOCI) indexing for SageMaker AI Training and processing jobs, enabling lazy loading of container images so jobs start in seconds instead of waiting for full image downloads. The feature works by fetching only needed files at launch and streaming the rest in the background, with benchmarks showing download time reductions of roughly 8x to 46x depending on image size. Users can adopt it via pre-indexed AWS Deep Learning Containers or by creating indexes for custom images, with no changes to job definitions required.

Amazon SageMaker AIcontainer image optimizationAI infrastructure
Industry Editor selected

OpenAI to stop training AI models on the live internet

OpenAI says it will no longer train its AI models on the live internet, citing "significant risks." The reported policy change comes weeks after one of its engineers said that approach wouldn't work. The source provides no further details on how training data will be sourced going forward.

Why it matters: It signals a notable shift in how a leading frontier lab plans to source training data, which matters to anyone tracking model development practices.

OpenAImodel trainingAI safety