Expert articles on RPA, AI automation, and enterprise technology by Alexander Reinike-Kaiser.
Get notified when we publish new insights on AI and automation.
Researchers have introduced MerchantBench, a new benchmark that tests AI agents in a complex, 365-day e-commerce simulation. The study reveals a significant "coherence gap," with even advanced AI models achieving less than 30% of the financial success of human participants.
AskChem is a new claim-centered infrastructure that breaks down millions of chemistry papers into 2.4 million verifiable, atomic findings. This shift from document-based search to evidence-based synthesis allows scientists and AI agents to assemble cross-paper answers with unprecedented accuracy and speed.
Researchers have developed HiFi-UMI, a high-fidelity data capture system that allows robots to learn complex manipulation tasks entirely from human demonstrations without needing a physical robot during training. This breakthrough could drastically lower the barrier to deploying versatile AI-driven automation in real-world environments.
Researchers have unveiled AREX, a deep research agent that uses recursive self-improvement and autonomous context management to solve complex, multi-constraint problems. This breakthrough allows smaller AI models to outperform massive industry giants by systematically auditing and refining their own work.
Researchers have unveiled ABot-World-0, a groundbreaking AI world model capable of streaming 720p interactive video at 16 FPS on a single desktop GPU. This technology enables long-horizon, controllable virtual environments, bridging the gap between static video generation and real-time interactive simulation.
Researchers have developed LongStraw, a new execution stack that enables reinforcement learning for AI models with context lengths exceeding 2 million tokens. This innovation bridges the gap between long-form inference and training, allowing AI agents to learn from massive datasets on limited hardware budgets.
