All published AI Radar articles.

Researchers have identified and analyzed "Perceptual Judgment Bias" in multimodal large language models (MLLMs) when used as evaluators. This bias causes MLLMs to prioritize plausible textual narratives over perceptually accurate visual evidence. To address this, they introduced the Perceptually Perturbed Judgment Dataset and a unified training framework combining GRPO-based rewards with a batch-ranking objective, aiming to improve perceptual fidelity and alignment with human evaluation.

Researchers have introduced Staged Executable Inverse Graphics (SEIG), an agentic framework that reconstructs 3D scenes from single images directly into editable Blender programs using pretrained vision-language models (VLMs). This method refines scene elements like geometry, materials, and lighting in Blender code, demonstrating improved reconstruction fidelity through task decomposition.

Muapi has released Harrlogos-XL, a LoRA model on Hugging Face designed for custom text generation within Stable Diffusion XL. This model aims to enhance text-to-image capabilities, specifically for integrating custom text into generated images.

The Harness Starter Kit is a GitHub repository designed to make software repositories safer and more effective for AI coding agents. It introduces a "prompt-first" approach, treating the repository itself as the operating environment for agents, and includes tools like "Harness Doctor" to evaluate repository readiness.

Siclaw is an open-source, AI-powered SRE platform designed for read-only infrastructure diagnostics. It helps DevOps and SRE teams investigate problems, gather evidence, form hypotheses, and perform root cause analysis without directly altering the environment. The platform supports deep investigation workflows, learns from past incidents, and offers multi-channel access via terminal, web UI, or chat.

OpenAI has outlined its approach to AI policy and political advocacy, emphasizing transparency, support for thoughtful regulation, and AI safety. The company also clarified that no external political group represents its views.

Inngest has introduced Utah, an open-source, TypeScript-based personal AI agent harness. Designed as an OpenClaw-like agent, Utah leverages Inngest for durable execution, retries, and observability, and uses `pi-ai` for a unified LLM interface. It supports multi-channel communication and local development.

EvoScientist, a GitHub project focused on self-evolving AI scientists for research, has released version 0.1.1. This update includes an upgrade to DeepAgents 0.6.2 with DeltaChannel for improved message persistence, tier-aware virtual mount resolution for shell commands, a status bar with phase and elapsed time indicators, and inline keyboard buttons for QQ Bot HITL prompts. It also adds a new `dashscope-code` provider for Alibaba Bailian Coding Plan keys and fixes issues with `@file` mentions containing spaces.

Strands Agents has released documentation for its SDK, a framework designed for building AI agents using a model-driven approach. The documentation, hosted on GitHub, provides guides, examples, and API references for developers.

OpenAI has commenced construction on a 1-gigawatt (GW) data center in Michigan, a key component of its Stargate initiative. This project aims to expand AI infrastructure, create jobs, and support local communities.