Sign in to report inappropriate content. Perturbations that we apply to the real robot hand while it solves the Rubik's Cube. All videos play at real-time.

We've trained a pair of neural networks to solve the Rubik's Cube with a human-like robot hand. The neural networks are trained entirely in simulation, using the same reinforcement learning code as OpenAI Five paired with a new technique called Automatic Domain Randomization (ADR). The system can handle situations it never saw during training, such as being prodded by a stuffed giraffe. This shows that reinforcement learning isn't just a tool for virtual tasks, but can solve physical-world problems requiring unprecedented dexterity. Human hands let us solve a wide variety of tasks. For the past 60 years of robotics, hard tasks which humans accomplish with their fixed pair of hands have required designing a custom robot for each task.

Sign in to report inappropriate content. We've trained a pair of neural networks to solve the Rubik's Cube with a human-like robot hand.

There's always been something so annoying about people who found the need to stack additional challenges onto solving a Rubik's Cube quickly, whether it was doing it blind-folded or while juggling or one-handed. While it might have just been a challenge for them, it also seemed like a need to show off. OpenAI is clearly interested in showing off what its Dactyl robotic-hand can do with a Rubik's Cube. The organization announced that the robot has learned to solve a Rubik's Cube one-handed, an accomplishment that speaks to the robot's dexterity in handling and manipulating the cube more than anything. Previously, we had seen the robot interact with unknown objects without any real-world training, only virtual simulations.

Robots with truly humanlike dexterity are far from becoming reality, but progress accelerated by AI has brought us closer to achieving this vision than ever before. In a research paper published in September, a team of scientists at Google detailed their tests with a robotic hand that enabled it to rotate Baoding balls with minimal training data. And at a computer vision conference in June, MIT researchers presented their work on an AI model capable of predicting the tactility of physical things from snippets of visual data alone. Now, OpenAI -- the San Francisco-based AI research firm cofounded by Elon Musk and others, with backing from luminaries like LinkedIn cofounder Reid Hoffman and former Y Combinator president Sam Altman -- says it's on the cusp of solving something of a grand challenge in robotics and AI systems: solving a Rubik's cube. Unlike breakthroughs achieved by teams at the University of California, Irvine and elsewhere, which leveraged machines tailor-built to manipulate Rubik's cubes with speed, the approach devised by OpenAI researchers uses a five-fingered humanoid hand guided by an AI model with 13,000 years of cumulative experience -- on the same order of magnitude as the 40,000 years used by OpenAI's Dota-playing bot.

To coordinate actions with an interaction partner requires a constant exchange of sensorimotor signals. Humans acquire these skills in infancy and early childhood mostly by imitation learning and active engagement with a skilled partner. They require the ability to predict and adapt to one's partner during an interaction. In this work we want to explore these ideas in a human-robot interaction setting in which a robot is required to learn interactive tasks from a combination of observational and kinesthetic learning. To this end, we propose a deep learning framework consisting of a number of components for (1) human and robot motion embedding, (2) motion prediction of the human partner and (3) generation of robot joint trajectories matching the human motion. To test these ideas, we collect human-human interaction data and human-robot interaction data of four interactive tasks "hand-shake", "hand-wave", "parachute fist-bump" and "rocket fist-bump". We demonstrate experimentally the importance of predictive and adaptive components as well as low-level abstractions to successfully learn to imitate human behavior in interactive social tasks.

It's (mostly) powered by OpenAI's GPT-2 and has additional fine-tuned models: The main technical challenges were creating an app that could deliver OpenAI's GPT-2 Medium (a ml model that generates text) quickly and simultaneously support 10-20 heavy users. Above steps had to be fully automated or scaling failed. It feels dirty (in 2019) to write bash scripts as part of an automated deployment, but it's necessary when using autoscaling on Google startup-scripts. Kubernetes is another option, but I'm not smart enough to use K8. Google startup-scripts runs a shell script when the machine starts.

Last year, Elon Musk praised the work of OpenAI after a team of five neural networks had defeated five humans, who ranked in the top 99.95 percentile of players worldwide, in the popular game Dota 2. The five bots had learned the game by playing against itself at a rate of a staggering 180 years per day. The game requires strong teamwork among the five players and, therefore, the achievement is quite remarkable and more evidence that artificial intelligence (AI) is rapidly becoming more advanced. However, directly after the five bots beat the five humans 2–1, Musk cautioned for the power of AI by urging that OpenAI should focus on AI that works with humans, instead of against humans. His statement is in line with his previous warnings for AI, which Musk believes could result in a robot dictatorship or an AI-arms race amongst superpowers that could be the most plausible cause for World War III. With artificial intelligence becoming increasingly sophisticated, also the warnings against AI become more pervasive, and the question remains then, is AI good or bad?

On Valentine's Day, OpenAI gifted us a paper – Better Language Models and Their Implications – that rocked my educator's world. OpenAI had developed an artificial intelligence (AI) model that had learnt, in an unsupervised way using millions of webpages, how to undertake writing tasks, many of which were of reasonable quality according to objective benchmarks. Imagine a future where an AI responds to an assessment task by producing original writing at pass or credit levels. No two responses would be the same because the AI would learn to check against what it and other AI had already produced. Traditional written assessment relies on students producing original work.

That said, I am really confused by OpenAI's PR strategy for GPT-2; they push it heavily as the future of text writing, but all of their GPT-2 related repos are archived w/o any updates expected.