Developing
Procgen and MineRL Competitions
OpenAI collaborates with leading institutions to host Procgen and MineRL competitions, pushing the boundaries of AI in gaming environments.
OpenAI NewsOriginal →
News
2,468 stories
New Release
Image GPT
OpenAI's Image GPT merges text and image generation, setting a new standard in creative AI applications.
OpenAI NewsOriginal →
New Release
OpenAI API
OpenAI unveils a new API, offering developers easy access to advanced AI models and enhanced performance.
OpenAI NewsOriginal →
New Release
Language models are few-shot learners
OpenAI's latest findings reveal that language models can learn tasks with minimal examples, revolutionizing AI efficiency.
OpenAI NewsOriginal →
New Release
AI and efficiency
AI training efficiency has surged, slashing compute needs by 44 times since 2012, revolutionizing model development.
OpenAI NewsOriginal →
New Release
Jukebox
OpenAI's Jukebox launches, enabling AI-generated music and singing across diverse genres.
OpenAI NewsOriginal →
New Release
Improving verifiability in AI development
OpenAI releases a report detailing ten mechanisms to enhance verifiability in AI development.
OpenAI NewsOriginal →
New Release
OpenAI Microscope
OpenAI Microscope launches to enhance neural network interpretability with detailed visualizations.
OpenAI NewsOriginal →
New Release
How to generate text: using different decoding methods for language generation with Transformers
Hugging Face unveils decoding methods to enhance text generation with Transformers, improving output quality and efficiency.
Hugging Face BlogOriginal →
New Release
How to train a new language model from scratch using Transformers and Tokenizers
Unlock the potential of custom language models with Hugging Face's latest guide on training from scratch.
Hugging Face BlogOriginal →
New Release
OpenAI standardizes on PyTorch
OpenAI adopts PyTorch as its standard deep learning framework to boost collaboration and efficiency.
OpenAI NewsOriginal →
New Release
Scaling laws for neural language models
New research uncovers scaling laws that enhance the performance of neural language models, guiding developers in model selection.
OpenAI NewsOriginal →
New Release
Dota 2 with large scale deep reinforcement learning
Deep reinforcement learning models are changing the game in Dota 2, outperforming human players with advanced strategies.
OpenAI NewsOriginal →
New Release
Deep double descent
New research uncovers the double descent phenomenon in neural networks, impacting model training and performance.
OpenAI NewsOriginal →
New Release
Procgen Benchmark
OpenAI launches Procgen Benchmark, a new tool for evaluating reinforcement learning agents across 16 unique environments.
OpenAI NewsOriginal →
New Release
Safety Gym
OpenAI unveils Safety Gym, a suite designed to enhance safety in reinforcement learning environments.
OpenAI NewsOriginal →
New Release
Benchmarking safe exploration in deep reinforcement learning
OpenAI unveils new benchmarks aimed at enhancing safety in deep reinforcement learning models.
OpenAI NewsOriginal →
New Release
GPT-2: 1.5B release
OpenAI has officially released the final version of GPT-2, featuring 1.5 billion parameters and tools for output detection.
OpenAI NewsOriginal →
New Release
Solving Rubik’s Cube with a robot hand
An AI-trained robot hand showcases remarkable dexterity by solving a Rubik's Cube, even under unexpected conditions.
OpenAI NewsOriginal →
New Release
OpenAI Scholars 2020: Applications open
OpenAI opens applications for its third class of Scholars, aiming to diversify the AI field.
OpenAI NewsOriginal →
New Release
Fine-tuning GPT-2 from human preferences
OpenAI's fine-tuning of GPT-2 with human feedback boosts performance and user alignment in summarization tasks.
OpenAI NewsOriginal →
New Release
Emergent tool use from multi-agent interaction
AI agents showcase advanced strategies and tool use in a hide-and-seek game, revealing new insights into collaborative learning.
OpenAI NewsOriginal →
Developing
Testing robustness against unforeseen adversaries
OpenAI introduces UAR, a new metric to enhance neural network defenses against unforeseen adversarial attacks.
OpenAI NewsOriginal →