If you’ve ever wondered: What is OpenAI?, you’re not alone. This non-profit organization was founded in 2014, and its mission is to develop kunstliche intelligence, or KI, on an open source basis that benefits society and its members. As such, OpenAI makes its research and patents publicly available. And it has already achieved several milestones.
One of its achievements was beating the world champion team in Dota 2, an extremely complex online game with its own E-Sports league. Other notable achievements include automated text recognition and image classification.
While Elon Musk calls kunstliche intelligence a threat to humanity, many other leading scientists and technologists are equally concerned about the ramifications of this technology. Many in related fields worry about the negative consequences of KI, and OpenAI aims to mitigate them. But what is KI, and how can it be used to improve our lives? Let’s take a look. A human-level KI would be a “human extension” of humans, and would be an equal, vast class, according to OpenAI.
The first product from OpenAI is GPT-3, a speech-recognition system trained on billions of internet queries. This speech recognition system aims to answer questions in natural language, translate between languages, and generate improvised texts. But a big question is: how does OpenAI’s speech recognition system compare to human-level speech recognition? To answer that question, we need to know what we’re talking about! The OpenAI GPT-3 team is currently developing the next generation of this product and is working hard to improve its capabilities.
The GPT-2 model has been trained to write autonomous texts by using an eight million website dataset. This is an impressive achievement for a machine-learning model. However, it has several limitations. While most machine learning models are narrow, this new model is much more flexible. DeepMind’s AlphaGo program, for instance, can beat Go-Weltmeister, a human-like AI. The GPT-2 model can beat Tic Tac Toe!
The OpenAI Gym is a free software toolkit created by the Non-Profit Organisation OpenAI. It allows people to use Reinforcement Learning algorithms and compare them with each other. OpenAI Gym also has various simulation environments to test the AI agent. With OpenAI Gym, you can develop and test your algorithm with different ML frameworks and libraries. If you’re thinking about building an AI agent, check out OpenAI Gym first.
MuseNet, a project of OpenAI, is a deep neural network that trains itself to predict MIDI music notes. It’s also trained to play games against humans. Its first public demonstration of this technology came on 27 April when Dendi defeated the OpenAI Five bot in a live match. The CTO, Greg Brockman, said that the bot learned to play against itself and then played against humans. The bot, which was called MuseNet, has already learned to play ten instruments and ten different styles. As a result, it has a fairly complicated repertoire and can handle complex tasks.
OpenAI’s aim is to build the most flexible AI language system in the world. While many companies have tuned AI systems to handle a limited set of tasks, the OpenAI API can handle jobs similar to those of specialized systems. It is currently in a limited test phase, but will soon be made available to other companies for use in a variety of applications including games, customer support chat systems, and education products. So, what’s the secret to OpenAI’s success?
Its creators have made the MIT-license open, and OpenAI has translated it into a german-speaking version. Although it is not legally binding, this inoffical translation is intended to help deutschspeaking readers better understand the MIT-license. However, OpenAI has no claim to the content created with GPT-2, and instead asks users to responsibly use it, and to clearly indicate when it has been created with it.