|Founded||December 11, 2015|
|Headquarters||Pioneer Building, San Francisco, California, US|
|Products||DALL-E, GPT-3, GPT-2, OpenAI Gym|
Number of employees
|>120 (as of 2020[update])|
OpenAI is an artificial intelligence (AI) research laboratory consisting of the for-profit corporation OpenAI LP and its parent company, the non-profit OpenAI Inc. The company, considered a competitor to DeepMind, conducts research in the field of AI with the stated goal of promoting and developing friendly AI in a way that benefits humanity as a whole. The organization was founded in San Francisco in late 2015 by Elon Musk, Sam Altman, and others, who collectively pledged US$1 billion. Musk resigned from the board in February 2018 but remained a donor. In 2019, OpenAI LP received a US$1 billion investment from Microsoft.
In October 2015, Elon Musk, Sam Altman, and other investors announced the formation of OpenAI and pledged over US$1 billion to the venture. The organization stated they would "freely collaborate" with other institutions and researchers by making its patents and research open to the public.
On December 5, 2016, OpenAI released Universe, a software platform for measuring and training an AI's general intelligence across the world's supply of games, websites and other applications.
In 2019, OpenAI transitioned from non-profit to for-profit. The company distributed equity to its employees and partnered with Microsoft Corporation, who announced an investment package of US$1 billion into the company. OpenAI then announced its intention to commercially license its technologies, with Microsoft as its preferred partner.
In June 2020, OpenAI announced GPT-3, a language model trained on trillions of words from the Internet. It also announced that an associated API, named simply "the API", would form the heart of its first commercial product. GPT-3 is aimed at natural language answering of questions, but it can also translate between languages and coherently generate improvised text.
Other backers of the project include:
The group started in early January 2016 with nine researchers. According to Wired, Brockman met with Yoshua Bengio, one of the "founding fathers" of the deep learning movement, and drew up a list of the "best researchers in the field". Microsoft's Peter Lee stated that the cost of a top AI researcher exceeds the cost of a top NFL quarterback prospect. While OpenAI pays corporate-level (rather than nonprofit-level) salaries, it doesn't currently pay AI researchers salaries comparable to those of Facebook or Google. Nevertheless, Sutskever stated that he was willing to leave Google for OpenAI "partly of because of the very strong group of people and, to a very large extent, because of its mission." Brockman stated that "the best thing that I could imagine doing was moving humanity closer to building real AI in a safe way." OpenAI researcher Wojciech Zaremba stated that he turned down "borderline crazy" offers of two to three times his market value to join OpenAI instead.
Some scientists, such as Stephen Hawking and Stuart Russell, have articulated concerns that if advanced AI someday gains the ability to re-design itself at an ever-increasing rate, an unstoppable "intelligence explosion" could lead to human extinction. Musk characterizes AI as humanity's "biggest existential threat." OpenAI's founders structured it as a non-profit so that they could focus its research on creating a positive long-term human impact.
Musk and Altman have stated they are motivated in part by concerns about the existential risk from artificial general intelligence. OpenAI states that "it's hard to fathom how much human-level AI could benefit society," and that it is equally difficult to comprehend "how much it could damage society if built or used incorrectly". Research on safety cannot safely be postponed: "because of AI's surprising history, it's hard to predict when human-level AI might come within reach." OpenAI states that AI "should be an extension of individual human wills and, in the spirit of liberty, as broadly and evenly distributed as possible...", and which sentiment has been expressed elsewhere in reference to a potentially enormous class of AI-enabled products: "Are we really willing to let our society be infiltrated by autonomous software and hardware agents whose details of operation are known only to a select few? Of course not." Co-chair Sam Altman expects the decades-long project to surpass human intelligence.
Vishal Sikka, former CEO of Infosys, stated that an "openness" where the endeavor would "produce results generally in the greater interest of humanity" was a fundamental requirement for his support, and that OpenAI "aligns very nicely with our long-held values" and their "endeavor to do purposeful work". Cade Metz of Wired suggests that corporations such as Amazon may be motivated by a desire to use open-source software and data to level the playing field against corporations such as Google and Facebook that own enormous supplies of proprietary data. Altman states that Y Combinator companies will share their data with OpenAI.
In 2019, OpenAI became a for profit company called OpenAI LP to secure additional funding while staying controlled by a non-profit called OpenAI Inc in a structure that OpenAI calls "capped-profit", having previously been a 501(c)(3) nonprofit organization.
Musk posed the question: "what is the best thing we can do to ensure the future is good? We could sit on the sidelines or we can encourage regulatory oversight, or we could participate with the right structure with people who care deeply about developing AI in a way that is safe and is beneficial to humanity." Musk acknowledged that "there is always some risk that in actually trying to advance (friendly) AI we may create the thing we are concerned about"; nonetheless, the best defense is "to empower as many people as possible to have AI. If everyone has AI powers, then there's not any one person or a small set of individuals who can have AI superpower."
Musk and Altman's counter-intuitive strategy of trying to reduce the risk that AI will cause overall harm, by giving AI to everyone, is controversial among those who are concerned with existential risk from artificial intelligence. Philosopher Nick Bostrom is skeptical of Musk's approach: "If you have a button that could do bad things to the world, you don't want to give it to everyone." During a 2016 conversation about the technological singularity, Altman said that "we don't plan to release all of our source code" and mentioned a plan to "allow wide swaths of the world to elect representatives to a new governance board". Greg Brockman stated that "Our goal right now... is to do the best thing there is to do. It's a little vague."
Conversely, OpenAI's initial decision to withhold GPT-2 due to a wish to "err on the side of caution" in the presence of potential misuse, has been criticized by advocates of openness. Delip Rao, an expert in text generation, stated "I don't think [OpenAI] spent enough time proving [GPT-2] was actually dangerous." Other critics argued that open publication is necessary to replicate the research and to be able to come up with countermeasures.
In the 2017 tax year, OpenAI spent US$7.9 million, or a quarter of its functional expenses, on cloud computing alone. In comparison, DeepMind's total expenses in 2017 were much larger, measuring US$442 million. In Summer 2018, simply training OpenAI's Dota 2 bots required renting 128,000 CPUs and 256 GPUs from Google for multiple weeks. According to OpenAI, the capped-profit model adopted in March 2019 allows OpenAI LP to legally attract investment from venture funds, and in addition, to grant employees stakes in the company, the goal being that they can say "I'm going to Open AI, but in the long term it's not going to be disadvantageous to us as a family." Many top researchers work for Google Brain, DeepMind, or Facebook, Inc., which offer stock options that a nonprofit would be unable to. In June 2019, OpenAI LP raised a billion dollars from Microsoft, a sum which OpenAI plans to have spent "within five years, and possibly much faster". Altman has stated that even a billion dollars may turn out to be insufficient, and that the lab may ultimately need "more capital than any non-profit has ever raised" to achieve AGI.
The transition from a nonprofit to a capped-profit company was viewed with skepticism by Oren Etzioni of the nonprofit Allen Institute for AI, who agreed that wooing top researchers to a nonprofit is difficult, but stated "I disagree with the notion that a nonprofit can't compete" and pointed to successful low-budget projects by OpenAI and others. "If bigger and better funded was always better, then IBM would still be number one." Following the transition, public disclosure of the compensation of top employees at OpenAI LP is no longer legally required. The nonprofit, OpenAI Inc., is the sole controlling shareholder of OpenAI LP. OpenAI LP, despite being a for-profit company, retains a formal fiduciary responsibility to OpenAI's Inc.'s nonprofit charter. A majority of OpenAI Inc.'s board is barred from having financial stakes in OpenAI LP. In addition, minority members with a stake in OpenAI LP are barred from certain votes due to conflict of interest. Some researchers have argued that OpenAI LP's switch to for-profit status is inconsistent with OpenAI's claims to be "democratizing" AI. A journalist in Vice News wrote that "generally, we've never been able to rely on venture capitalists to better humanity".
Gym aims to provide an easy to set up, general-intelligence benchmark with a wide variety of different environments—somewhat akin to, but broader than, the ImageNet Large Scale Visual Recognition Challenge used in supervised learning research—and that hopes to standardize the way in which environments are defined in AI research publications, so that published research becomes more easily reproducible. The project claims to provide the user with a simple interface. As of June 2017, Gym can only be used with Python. As of September 2017, the Gym documentation site was not maintained, and active work focused instead on its GitHub page.
In "RoboSumo", virtual humanoid "metalearning" robots initially lack knowledge of how to even walk, and given the goals of learning to move around, and pushing the opposing agent out of the ring. Through this adversarial learning process, the agents learn how to adapt to changing conditions; when an agent is then removed from this virtual environment and placed in a new virtual environment with high winds, the agent braces to remain upright, suggesting it had learned how to balance in a generalized way. OpenAI's Igor Mordatch argues that competition between agents can create an intelligence "arms race" that can increase an agent's ability to function, even outside the context of the competition.
In 2018, OpenAI launched the Debate Game, which teaches machines to debate toy problems in front of a human judge. The purpose is to research whether such an approach may assist in auditing AI decisions and in developing explainable AI.
Dactyl uses machine learning to train a robot Shadow Hand from scratch, using the same reinforcement learning algorithm code that OpenAI Five uses. The robot hand is trained entirely in physically inaccurate simulation.
The original paper on generative pre-training (GPT) of a language model was written by Alec Radford and colleagues, and published in preprint on OpenAI's website on June 11, 2018. It showed how a generative model of language is able to acquire world knowledge and process long-range dependencies by pre-training on a diverse corpus with long stretches of contiguous text.
Generative Pre-trained Transformer 2, commonly known by its abbreviated form GPT-2, is an unsupervised transformer language model and the successor to GPT. GPT-2 was first announced in February 2019, with only limited demonstrative versions initially released to the public. The full version of GPT-2 was not immediately released out of concern over potential misuse, including applications for writing fake news. Some experts expressed skepticism that GPT-2 posed a significant threat. The Allen Institute for Artificial Intelligence responded to GPT-2 with a tool to detect "neural fake news". Other researchers, such as Jeremy Howard, warned of "the technology to totally fill Twitter, email, and the web up with reasonable-sounding, context-appropriate prose, which would drown out all other speech and be impossible to filter". In November 2019, OpenAI released the complete version of the GPT-2 language model. Several websites host interactive demonstrations of different instances of GPT-2 and other transformer models.
GPT-2's authors argue unsupervised language models to be general-purpose learners, illustrated by GPT-2 achieving state-of-the-art accuracy and perplexity on 7 of 8 zero-shot tasks (i.e. the model was not further trained on any task-specific input-output examples). The corpus it was trained on, called WebText, contains slightly over 8 million documents for a total of 40 GB of text from URLs shared in Reddit submissions with at least 3 upvotes. It avoids certain issues encoding vocabulary with word tokens by using byte pair encoding. This allows to represent any string of characters by encoding both individual characters and multiple-character tokens.
Generative Pre-trained[a] Transformer 3, commonly known by its abbreviated form GPT-3, is an unsupervised Transformer language model and the successor to GPT-2. It was first described in May 2020. OpenAI stated that full version of GPT-3 contains 175 billion parameters, two orders of magnitude larger than the 1.5 billion parameters in the full version of GPT-2 (although GPT-3 models with as few as 125 million parameters were also trained).
OpenAI stated that GPT-3 succeeds at certain "meta-learning" tasks. It can generalize the purpose of a single input-output pair. The paper gives an example of translation and cross-linguistic transfer learning between English and Romanian, and between English and German.
GPT-3 dramatically improved benchmark results over GPT-2. OpenAI cautioned that such scaling up of language models could be approaching or encountering the fundamental capability limitations of predictive language models. Pre-training GPT-3 required several thousand petaflop/s-days[b] of compute, compared to tens of petaflop/s-days for the full GPT-2 model. Like that of its predecessor, GPT-3's fully trained model was not immediately released to the public on the grounds of possible abuse, though OpenAI planned to allow access through a paid cloud API after a two-month free private beta that began in June 2020.
OpenAI's MuseNet (2019) is a deep neural net trained to predict subsequent musical notes in MIDI music files. It can generate songs with ten different instruments in fifteen different styles. According to The Verge, a song generated by MuseNet tends to start out reasonably but then fall into chaos the longer it plays.
OpenAI's Jukebox (2020) is an open-sourced algorithm to generate music with vocals. After training on 1.2 million samples, the system accepts a genre, artist, and a snippet of lyrics, and outputs song samples. OpenAI stated the songs "show local musical coherence, follow traditional chord patterns" but acknowledged that the songs lack "familiar larger musical structures such as choruses that repeat" and that "there is a significant gap" between Jukebox and human-generated music. The Verge stated "It's technologically impressive, even if the results sound like mushy versions of songs that might feel familiar", while Business Insider stated "surprisingly, some of the resulting songs are catchy and sound legitimate".
DALL-E is a Transformer model that creates images from textual descriptions, revealed by OpenAI in January 2021.
CLIP does the opposite: it creates a description for a given image. DALL-E uses a 12-billion-parameter version of GPT-3 to interpret natural language inputs (such as "a green leather purse shaped like a pentagon" or "an isometric view of a sad capybara") and generate corresponding images. It is able to create images of realistic objects ("a stained glass window with an image of a blue strawberry") as well as objects that do not exist in reality ("a cube with the texture of a porcupine"). As of March 2021, no API or code is available.
In March 2021 OpenAI released a paper, titled Multimodal Neurons in Artificial Neural Networks, where they showed detailed analysis of CLIP (and GPT) models and their vulnerabilities. The new type of attacks on such models was described in this work.
We refer to these attacks as typographic attacks. We believe attacks such as those described above are far from simply an academic concern. By exploiting the model’s ability to read text robustly, we find that even photographs of hand-written text can often fool the model.— Multimodal Neurons in Artificial Neural Networks, OpenAI
OpenAI Microscope is a collection of visualizations of every significant layer and neuron of eight different neural network models which are often studied in interpretability. Microscope was created for easy analysis of the features that form inside these neural networks. The models included are AlexNet, VGG 19, different versions of Inception, and different versions of CLIP Resnet.
OpenAI Codex is a descendant of GPT-3 that has additionally been trained on code from 54 million GitHub repositories. It was announced in mid-2021 as the AI powering the code autocompletion tool GitHub Copilot. In August 2021, an API was released in private beta. According to OpenAI, the model is able to create working code in over a dozen programming languages, most effectively in Python.
OpenAI Five is the name of a team of five OpenAI-curated bots that are used in the competitive five-on-five video game Dota 2, who learn to play against human players at a high skill level entirely through trial-and-error algorithms. Before becoming a team of five, the first public demonstration occurred at The International 2017, the annual premiere championship tournament for the game, where Dendi, a professional Ukrainian player, lost against a bot in a live 1v1 matchup. After the match, CTO Greg Brockman explained that the bot had learned by playing against itself for two weeks of real time, and that the learning software was a step in the direction of creating software that can handle complex tasks like a surgeon. The system uses a form of reinforcement learning, as the bots learn over time by playing against themselves hundreds of times a day for months, and are rewarded for actions such as killing an enemy and taking map objectives.
By June 2018, the ability of the bots expanded to play together as a full team of five and they were able to defeat teams of amateur and semi-professional players. At The International 2018, OpenAI Five played in two exhibition matches against professional players, but ended up losing both games. In April 2019, OpenAI Five defeated OG, the reigning world champions of the game at the time, 2:0 in a live exhibition match in San Francisco. The bots' final public appearance came later that month, where they played in 42,729 total games in a four-day open online competition, winning 99.4% of those games.
Gym Retro is a platform for reinforcement learning research on games. Gym Retro is used to conduct research on RL algorithms and study generalization. Prior research in RL has mostly focused on optimizing agents to solve single tasks. Gym Retro gives the ability to generalize between games with similar concepts but different appearances.
Altman said they expect this decades-long project to surpass human intelligence.
Elon Musk: ...we came to the conclusion that having a 501(c)(3)... would probably be a good thing to do
The intuition behind pre-trained language models is to create a black box which understands the language and can then be asked to do any specific task in that language.
GPT-2, is a 1.5B parameter TransformerCite journal requires
Since we increase the capacity by over two orders of magnitude from GPT-2 to GPT-3
A petaflop/s-day (pfs-day) consists of performing 1015 neural net operations per second for one day, or a total of about 1020 operations. The compute-time product serves as a mental convenience, similar to kW-hr for energy.
Why did OpenAI choose to release an API instead of open-sourcing the models?
There are three main reasons we did this. First, commercializing the technology helps us pay for our ongoing AI research, safety, and policy efforts. Second, many of the models underlying the API are very large, taking a lot of expertise to develop and deploy and making them very expensive to run. This makes it hard for anyone except larger companies to benefit from the underlying technology. We’re hopeful that the API will make powerful AI systems more accessible to smaller businesses and organizations. Third, the API model allows us to more easily respond to misuse of the technology. Since it is hard to predict the downstream use cases of our models, it feels inherently safer to release them via an API and broaden access over time, rather than release an open source model where access cannot be adjusted if it turns out to have harmful applications.
If you’ve ever wanted to try out OpenAI’s vaunted machine learning toolset, it just got a lot easier. The company has released an API that lets developers call its AI tools in on “virtually any English language task.”