Articles & Guides

DataPrep-Bench evaluates how well LLMs, agents, and data workflows prepare training data by measuring downstream utility rather than surface-level text quality. It jointly benchmarks data construction and data quality evaluation across domains and base models.

IDEAgent tackles a core weakness in LLM-based research ideation: systems often optimize either quality or diversity, but not both together. The paper introduces a multi-agent framework and a Yield metric to measure diverse, quality-qualified idea sets.

Sony researchers introduce Spectral Alignment, a lightweight inference-time guidance method for correcting frequency-dependent mismatch in diffusion and flow-matching models. The method calibrates intermediate predictions with a precomputed spectral prior while adding only modest overhead.

Three-Body Scattering Modeling offers a new training signal for one-step generators by learning how generated samples should move toward real data, without an adversarial critic or a prescribed diffusion path.

The paper introduces Agentic Context Management, arguing that production agent failures often stem from poor control over what sits in the reasoning context, not from weak reasoning alone.

Netflix’s ID-V2V targets identity-preserving video restylization: propagate scene, lighting, and style edits from keyframes across a source video without losing facial likeness, expressions, gaze, or lip sync.

Microsoft has introduced MAI-Cyber-1-Flash and Project Perception, two AI-driven security tools designed to find, assess, and help fix vulnerabilities. The company claims benchmark and cost advantages, yet the timing highlights unresolved concerns around autonomous security agents.

Microsoft CEO Satya Nadella argues that companies relying wholly on proprietary AI labs risk outsourcing not just infrastructure, but their own thinking. His proposed answer is to keep prompts, metadata, context, memory, and agent tooling under enterprise control.

Some Claude shared chats and Artifacts were reportedly searchable on Google, with examples said to include sensitive health, workplace and children’s contact information. The incident highlights the risks of treating “anyone with the link” as private.

Google’s attempt to use the DMCA to stop SerpApi from scraping search results has suffered an early court defeat. The case highlights how AI-era scraping disputes are pushing copyright law into unfamiliar territory.

Microsoft has introduced MAI-Cyber-1-Flash and Perception, a new agentic cybersecurity platform designed to help enterprises find, triage, and remediate vulnerabilities faster.

ChatGPT is reportedly declining prompts that ask it to write in the exact style of well-known authors. Instead, it offers to capture broader qualities or a similar mood while keeping the output distinct.

New Similarweb data cited by TechCrunch shows Google’s AI Overviews now appear in 43% of searches, up from 15% a year earlier. The shift suggests AI-generated answers are becoming central to how people find information online.

Safe Superintelligence, the AI lab founded by Ilya Sutskever, has emerged from two quiet years with a long-term Nvidia partnership. The deal gives SSI access to Nvidia’s Vera Rubin GPU platform and signals a major scale-up for its safety-focused research agenda.

AMD’s new MI455X accelerator and Helios rack system show that the company is no longer competing only at the GPU-card level. It is trying to sell a full AI infrastructure story built around advanced chiplets, HBM4 memory and open interconnects.

Cursor’s engineering team asked a swarm of agents to rebuild a SQLite-like engine in Rust using only the 835-page manual. The striking result was not just test performance, but the cost gap between different agent architectures.

NVIDIA has introduced Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics. The system distills Cosmos-H-Surgical-Simulator into a causal student model and serves it with the FlashDreams streaming inference stack.

AWS has released Loom in AWS Labs as an open-source reference platform for building and governing AI agents at enterprise scale. The project focuses on identity propagation, access control, deployment governance, agent registries, and human approval flows.

A Tencent Hunyuan paper studies how to scale native multimodal pre-training from scratch under a fixed compute budget. The work highlights that language and multimodal objectives follow different allocation behaviors.

Skill Self-Play proposes a way to make LLM self-evolution more reliable by placing verifiable agent skills at the center of the training loop. The framework co-evolves a task proposer, a solver, and a dynamic skill controller through reinforcement learning.

Molt is an open-source training framework designed to reduce the engineering overhead of agentic reinforcement learning research. It keeps the codebase compact while supporting asynchronous training for multimodal and mixture-of-experts policies.