AI
Introducing Activation Atlases
OpenAI has developed activation atlases, a new technique for visualizing interactions between neurons in AI systems. This collaboration with Google re...
AI
AI
Neural MMO: A massively multiagent game environment
OpenAI has released Neural MMO, a game environment designed for training reinforcement learning agents in complex scenarios. This platform allows for ...
AI
Spinning Up in Deep RL: Workshop review
OpenAI hosted a workshop on deep reinforcement learning (RL) called 'Spinning Up in Deep RL', which is part of their education initiative. The event a...
AI
AI safety needs social scientists
A recent paper argues that long-term AI safety research requires the involvement of social scientists. The authors claim that current alignment algori...
AI
Better language models and their implications
OpenAI has developed a large-scale unsupervised language model that can generate coherent paragraphs of text. The model achieves state-of-the-art perf...
AI
Computational limitations in robust classification and win-win results
Researchers at OpenAI have found that computational limitations can lead to improved performance in robust classification tasks. They discovered a 'wi...
AI
Generalized Language Models
The article discusses the advancements in natural language processing (NLP) in 2018. Large-scale pre-trained language models like OpenAI GPT and BERT ...
AI
Object Detection Part 4: Fast Detection Models
This article is part of a series on object recognition. The previous installment covered region-based object detection algorithms that are accurate bu...
AI
OpenAI Fellows Summer 2018: Final projects
OpenAI's first cohort of fellows completed their six-month apprenticeships. The program aimed to turn machine learning beginners into core contributor...
AI
How AI training scales
Researchers have found a way to predict how well artificial intelligence (AI) can be trained in parallel, which could lead to faster and more efficien...
AI
Quantifying generalization in reinforcement learning
Researchers at OpenAI have developed a training environment called CoinRun to measure an agent's ability to generalize its experience in reinforcement...
AI
Meta-Learning: Learning to Learn Fast
Meta-Learning is a subfield of machine learning that involves training models to learn how to learn from data. This approach is useful for tasks where...