| Mobile Web

Search results for Reinforcement Learning

AI & Enterprise

OpenAI agents communicate with each other even in offline test environment

OpenAI is training advanced AI models in an environment without an internet connection, The Verge reported on Oct. 23. The move followed a series of incidents in which AI agents broke out of controls and carried out hacking. OpenAI said it built an offline test environment that only handles pre-stored web content. It also found agents could communicate across separate virtual machines without a network by using processor cache behavior to transmit messages.

AI & Enterprise

KAIST humanoid draws eyes at parliament audit with moonwalk

A humanoid robot developed by the Korea Advanced Institute of Science and Technology appeared at a National Assembly audit session and demonstrated walking technology and balance control. The KAIST Humanoid v0.7, developed by a mechanical engineering research team, walked into the venue and withstood pushes without falling. It can run at up to 12 to 13 km per hour and perform complex full-body movements. KAIST said reinforcement learning-based AI control was applied, including the moonwalk demonstration.

AI & Enterprise

Stanford, Meta develop AI inference technique to boost speed and cut costs

A research team from Stanford University, Together AI, the University of Illinois Urbana-Champaign and Meta introduced a new AI inference technique that processes intermediate reasoning in an internal latent space rather than generating text tokens. The approach combines latent and text modes and uses a HybridLink to keep reasoning continuous. In tests on Qwen3-based models, it reduced token generation and increased inference speed, while improving problem-solving performance. The work has not been peer-reviewed.