Why has OpenAI been so effective?

Article By : AI-STEVEN

I personally think they’ve been extremely effective because they’ve been able to attract some of the world’s best deep learning researchers, focus on some of the most central problems in AI, and create open source projects that further progress in the field. They’re able to do all of this while not having to deal with […]

I personally think they’ve been extremely effective because they’ve been able to attract some of the world’s best deep learning researchers, focus on some of the most central problems in AI, and create open source projects that further progress in the field. They’re able to do all of this while not having to deal with the demands of traditional companies in industry or groups in academia.

Quality of Researchers

Just to mention a couple names…

  • – Was part of the group that published the AlexNet paper (which some say was one of the driving forces behind the deep learning revolution).
  • Andrej Karpathy
  • – You may know him from his incredibly well written blog posts.
  • Alec Radford – Created DCGANs, which was an incredible breakthrough in image generation.

Not to mention the two big name cochairs of the project, Elon Musk (CEO of Tesla and SpaceX) and Sam Altman (President of Y Combinator).

Working on the Important Problems

Ever since their creation 17 months ago, it’s clear that OpenAI has largely focused on reinforcement learning. This is definitely a great direction to head in, especially since OpenAI’s mission is “Discovering and enacting the path to safe and general artificial intelligence”. Some of the most interesting work comes with the problems of effective agent communication

to one-shot imitation learning, OpenAI also has previously worked with creating defenses against adversarial examples as well as unsupervised learning

with natural language.

Open Source Projects

The two most influential projects that OpenAI created were Gym

and Universe (There’s also another project called Roboschool, which just got released last week!). These are two software platforms that allow users to create, train, and evaluate reinforcement learning algorithms and agents. These type of well made open source projects make it very easy for beginners and researchers around the world to prototype their ideas and to push the state of the art in RL.

Subscribe to Newsletter

Test Qr code text s ss