Keep pulling the thread on Jack Clark.
Using the ENPIRE framework, coding agents autonomously developed policies achieving a 99% success rate on dexterous manipulation tasks such as PushT, organizing pins, and cutting a zip tie.
Tencent has deployed its ARGUS software on a production cluster of over 10,000 GPUs for more than six months.
Tencent has conducted large-scale training runs, including a 4,096-GPU job for a HunyuanVideo model, a 512-GPU job for an audio model, and a 12,960-GPU job for a Hunyuan LLM Mixture-of-Experts model.
States that remove humans from the decision-making loop and delegate more control to AI will gain a strategic advantage in conflicts.
Human autonomy will be lost even with perfectly aligned AI because humanity will be governed by omniscient, omnipotent AI masters.
NVIDIA has developed ENPIRE, a software framework for physical robots to perform autonomous experimentation and execution.
The experimental hardware for NVIDIA's ENPIRE project includes two YAM (Yet Another Manipulator) arms from I2RT and an NVIDIA RTX 5090 GPU per workstation.
In NVIDIA's ENPIRE experiments, models based on GPT-5.5 and Opus 4.7 demonstrated the best performance, while Kimi-2.6 lagged behind.
Experiments with NVIDIA's ENPIRE framework show that using a larger number of agents, such as eight, leads to finding higher-scoring solutions more quickly.
A challenge in scaling the ENPIRE system is that Mean Robot Utilization (MRU) decreases as the number of robots increases, due to agents waiting for the language model.
Tencent developed the ARGUS software for telemetry and debugging errors in large-scale GPU clusters used for AI training.
Researchers at UC Berkeley have created the Local Ordinance Corpus for the United States (LOCUS), a dataset of U.S. municipal and county laws.