New Release
Learning from human preferences
OpenAI and DeepMind unveil a new algorithm that enhances AI safety by learning from human preferences.
OpenAI NewsOriginal →
2,472 stories
OpenAI and DeepMind unveil a new algorithm that enhances AI safety by learning from human preferences.
OpenAI explores multiagent environments to enhance AI adaptability and performance in pursuit of AGI.
OpenAI unveils Q-ensembles, a new method that enhances exploration strategies in reinforcement learning.
OpenAI launches DQN and its variants, enhancing the Baselines project for reinforcement learning developers.
New robots can master tasks after just one demonstration, revolutionizing training efficiency in various industries.
Roboschool launches as an open-source platform for robotic simulation, enhancing training capabilities for developers.
New research establishes a theoretical link between policy gradients and soft Q-learning, promising advancements in reinforcement learning efficiency.
OpenAI unveils stochastic neural networks to boost hierarchical reinforcement learning efficiency.
OpenAI's new unsupervised sentiment neuron revolutionizes sentiment analysis by leveraging Amazon reviews without labeled data.
OpenAI unveils groundbreaking AI that detects spam in the physical world using advanced simulation training.
Evolution strategies are emerging as a powerful alternative to traditional reinforcement learning techniques.
OpenAI introduces one-shot imitation learning, enabling AI to learn from a single example and transforming training efficiency.
OpenAI introduces Distill, a journal dedicated to enhancing clarity in machine learning research communication.
OpenAI's agents have developed a unique language, paving the way for advanced human-AI communication.
New research shows how multi-agent systems evolve language through interaction, paving the way for better AI collaboration.
OpenAI introduces temporal segment models that enhance prediction accuracy and control in dynamic AI environments.
OpenAI unveils third-person imitation learning, a breakthrough method to enhance AI training efficiency.
Adversarial examples threaten the integrity of machine learning models by exploiting their vulnerabilities, posing a significant security risk.
New research uncovers vulnerabilities in neural network policies, revealing weaknesses and proposing strategies for enhanced robustness.
OpenAI expands its team to 45 members, aiming to enhance AI capabilities and innovate in software systems.
OpenAI unveils PixelCNN++, a model that enhances image generation with advanced likelihood modeling techniques.
Poorly defined reward functions in reinforcement learning can lead to unexpected and harmful AI behaviors.
OpenAI launches Universe, a groundbreaking platform for measuring AI's general intelligence across various applications.