Meta's autonomous AI research system AIRA₃ won a Gold medal in a Kaggle competition, demonstrating human-expert-level model improvement capabilities.
World Labs co-founders discussed the world model Atlas and the future of spatial intelligence on the a16z podcast.
LLM-as-a-Verifier v0.2.0 has been released, achieving SOTA performance on agent benchmarks such as Terminal-Bench 2.1 and reducing uncached input tokens by approximately 3.4x through cache optimization.
OpenAI revealed its internal research automation status, stating that agent labor has surpassed human labor.
Salesforce has released Random Attention, which applies random sampling instead of signal-based methods for KV cache eviction in reasoning models.
xAI has released the Grok Imagine Video 1.5 agent, powered by its latest Image 2.0 model.
OpenAI's Chief Scientist proposed international coordination to slow down AI development, citing the limitations of alignment technology.
SemiAnalysis analyzed that under external buyer TCO and 100 tokens per second per user conditions for Qwen3.5 397B FP8 inference, Ironwood's cost per token is approximately 19% lower than B200.
OpenAI, WAN-IFRA, and AIRPPU have unveiled a collaborative program to enhance AI adoption and sustainability for independent media in Ukraine.
Version 2.2 has been released, offering deterministic control over task routing and verification processes for AI coding agents.
Developer Gustavo Arretureta has released 'Airuncode', a local AI agent runtime that supports multi-model debate and structural self-healing features.
Claude Code proved that the maximum number of lines for the 15-point Orchard Problem is 31, using SAT verification.
PINNStudio, a free open-source GUI that enables everything from PDE definition to training and visualization without complex coding for Physics-Informed Neural Networks (PINNs), has been released.
PyTorchKR selected this week's papers, introducing recent research on enhancing agent capabilities and interpreting internal structures, such as long-horizon autonomous coding agents, recursive self-improvement, and the approximation of symbolic structures in neural networks.
Armo's CTO proposed a behavior baseline-based anomaly detection solution to defend against Tool Output Injection.
A new technique for converting embeddings without paired data has revealed security vulnerabilities in vector databases.
An open-source CLI tool has been released to diagnose the MCP allowlist bypass issue in Claude Code.
OpenAI Chief Scientist Jakub Pachocki introduced alignment progress in GPT-6 Astra and called for international coordination and voluntary slowdowns in preparation for the era of recursive self-improvement.
AI Briefing