News
Google Warns Autonomous AI Agents Are Now Executing Full Cyberattacks
Google’s Threat Intelligence Group reports that adversaries have evolved from basic AI prompting to deploying autonomous agentic AI systems for…
Opinions
Ex-Anthropic Researcher Warns of AI ‘Catastrophe’ in Resignation
Jacob Coxon, a 27-year-old researcher who previously worked at Anthropic and OpenAI, resigned on September 8, warning that AI companies…
Opinions
AI Labs Face ‘Model Fatigue’ as Release Cycles Compress From 37 Days to 11
AI laboratories are under mounting pressure from accelerating release cycles, with the median interval between frontier model launches shrinking from…
Engineering
Environments as Scaffold: Enriching Feedback to Bootstrap Self-Evolving Agents in Long-Horizon Tasks
The paper introduces Feedback-Enriched Environments (FEEs), which shift training focus from agent-side improvements to environment-side adaptations rather than relying solely…
Engineering
Procedural Graphs: Self-Evolving Execution Structures for LLM Agents
The paper introduces Procedural Graphs, a framework that organizes procedural knowledge as structured (procedure, relation, procedure) triplets to guide LLM…
Engineering
NeoHorse-1: Recursive Self-Improvement via Agentic Post-Training With a Routing Harness
Researchers introduce NeoHorse-1, an approach to recursive self-improvement built on agentic post-training over a heterogeneous model pool with intelligent routing.…
Opinions
Generative AI Now Beats the Average Human on Creativity Tests, Study Finds
Researchers at the Université de Montréal conducted the largest comparative study of machine and human creativity, testing leading AI models…
Engineering
Online Draft Co-Training Speeds Up Speculative Decoding for Large-Scale RL Post-Training
Researchers present a system for accelerating reinforcement learning training by improving speculative decoding through online co-training of draft models. The…
Engineering
BeaconKV Compresses Key-Value Caches for Efficient Large Reasoning Model Inference
The paper proposes BeaconKV, a training-free technique that compresses key-value caches during inference for large reasoning models. It identifies that…
Engineering
Gander: An Omni Interaction Agent With a Brain-Cerebellum Architecture
Researchers present Gander, an end-to-end omni interaction agent that integrates multimodal perception, real-time conversation, and complex task execution within a…
Engineering
What LLM Trading Agents Actually Do in Production: A Six-Month, Population-Scale Study
Researchers analyzed 7.5 million single-model invocations and roughly 300,000 onchain actions across two deployed LLM trading-agent fleets over six months.…
Engineering
AlphaGenome Atlas: A Predictive Map of Every Possible DNA Letter Change in the Human Genome
Google DeepMind released AlphaGenome Atlas, a deep learning system that provides precomputed predictions for all 9 billion possible single-nucleotide variants…