OpenAI is making some big changes to its AI-powered coding assistant. The system now accepts plain English commands and outputs live, working code so that someone can easily build a game or web app without naming any variables at all. Coders (and those who don’t know how) will be able to try out this new API in the free private beta coming soon – hopefully it’ll make programming easier for everyone!
Codex is best thought of as a versatile language engine that was trained only on code instead of ordinary written material. Codex can now complete lines or entire sections, but when it was announced it wasn’t something non-coders would be able to interact with easily. That all changed with this new API which interprets simple requests like “make the ball bounce off the sides of the screen” and outputs working code in one dozen languages–all without any programming knowledge required.
Wojciech Zaremba (Codex lead) and Greg Brockman demonstrated how they were able to create a simple game from scratch. They explained the process behind selecting tools for their project, along with what OpenAI has in store for making AI more accessible in the near future.
Wojciech Zaremba and Greg Brockman live demoed creating a simple computer game from start to finish on stage at TED2018 this week using an open-source framework called TensorFlow-XG that is freely available online – demonstrating just how far machine learning advancements have come over recent years as well as some of OpenAI’s goals going forward like democratizing access to these technologies by working toward free versions across different platforms.
What is OpenAI?
OpenAI is a non-profit research company started by Elon Musk and Y Combinator’s Sam Altman in 2015. They are focused on advancing digital intelligence to make the world a better place. In that mission, they’ve open-sourced their work through their Github account.
Why is OpenAI important?
This week, OpenAI is unveiling an open-source deep learning software that will allow anyone to create their own AI bots – even if they don’t know how to program.
Their new library of tools allows coders and non-coders alike to create artificial agents that can learn from a human user and complete tasks on their behalf. The system is called AIX, or Artificial Intelligence eXplorer, and it’s meant for developers to build their own creations like games or programs. It could take non-programmers all the way from idea creation to product release without any actual coding required.
“It basically enables anyone that has an idea for a game to be able to make it in a way that they don’t have to be an expert coder,” Brockman tells me. “Anyone can put it together and share it with others.”
Brockman says the AIX is already capable of creating some impressive projects, like this Atari Breakout clone made entirely by OpenAI’s bots. He notes that figuring out how to make a robot move around in the game world was surprisingly challenging for their software and that this is just the first step toward AI learning how to create far more complex products.
The company also released an API as part of this project codenamed Codex which allows coders to use natural language commands like “make my character walk left” instead of having to describe what every single object in their creation looks like.
OpenAI’s tools are meant to be trained on a user’s own data and then work independently from there. The company hopes that doing so will encourage more people to create their own programs, instead of just relying on massive companies like Amazon who sell cloud services with pre-trained machine learning capabilities.
“I can’t compete with Amazon,” Brockman says. “They have access to all these resources that I just don’t have, but the idea of this is you could give an AI what it needs to start learning on its own and then continue doing so without having to pay for those cloud services.”
In addition to the tech and tools the company is releasing, OpenAI also has an impressive advisory board comprised of some of the brightest minds in AI. They include people like Yoshua Bengio from the University of Montreal (one of the world’s top experts on deep learning), as well as Fei-Fei Li from Stanford (known for her work with ImageNet). And the list goes on.
OpenAI’s ultimate goal is to advance AI while ensuring that it’s good for humans, and they seem to be doing just that by both launching tools that simply aren’t available anywhere else and also employing some of the world’s top minds in the field.
“We’re really trying to build a community of developers across a wide range of disciplines,” Brockman says. “The long-term goal is that these technologies will be available to everyone, but I also think this can help us build a community around how AI technology can best be used.”
At the end of the day, however, OpenAI’s tools are actually about making it easier for machine learning developers to build their own programs. Brockman says a lot of time in the field is spent on “crushing bugs,” and this system should free up people’s time for more important tasks like actually creating more advanced AI technology.
“I don’t think any of us have ever seen technology just crash and burn,” Brockman says. “But I’m really excited about this community we’re building.”