OpenAI facts for kids
.svg){kind=logo} |
|
| Private | |
| Industry | Artificial intelligence |
| Founded | December 11, 2015 |
| Headquarters | San Francisco, California, U.S. |
|
Key people
|
|
| Products | OpenAI Five |
| Revenue |  US$3.7 billion (2024 est.) |
 US$−5 billion (2024 est.) |
|
|
Number of employees
|
c. 2,000 (2024) |
OpenAI, Inc. is an American organization that researches and develops artificial intelligence (AI). It was started in December 2015 in San Francisco, California. Its main goal is to create "safe and helpful" artificial general intelligence (AGI). AGI means computer systems that can do most jobs better than humans.
OpenAI is a leader in the current AI boom. It is famous for its GPT family of large language models. These include DALL-E, which creates images from text, and Sora, which makes videos from text. When OpenAI released ChatGPT in November 2022, it made many people interested in generative AI.
The organization has two main parts. One is the non-profit OpenAI, Inc. The other is a for-profit company called OpenAI Global, LLC, which started in 2019. Their mission is to make sure AGI helps everyone. Microsoft owns about 49% of OpenAI and has invested $13 billion. Microsoft also provides powerful computers for OpenAI through its cloud platform, Microsoft Azure.
In 2023 and 2024, OpenAI faced some legal challenges. These were about using written works to train their AI products. In November 2023, the board of OpenAI made a big change. They removed Sam Altman as CEO, but he was back in his role just five days later. This happened after the board was reorganized. During 2024, some researchers who focused on AI safety left OpenAI. They pointed to concerns within the AI industry.
Contents
- History of OpenAI
- How OpenAI is Managed
- Why OpenAI Started
- OpenAI's Approach
- Products and Applications
- Learning Through Practice (Reinforcement Learning)
- API (Application Programming Interface)
- Creating Text (Text Generation)
- Image Understanding (Image Classification)
- Creating Images from Text (Text-to-Image)
- Creating Videos from Text (Text-to-Video)
- Turning Speech into Text (Speech-to-text)
- Creating Music (Music Generation)
- User Interfaces
- Stargate and Other Supercomputers
- See also
- Images for kids
History of OpenAI
Early Days as a Non-Profit (2015–2018)
_-1.jpg)
OpenAI was founded in December 2015. Key founders included Sam Altman, Elon Musk, Ilya Sutskever, and Greg Brockman. Sam Altman and Elon Musk were the co-chairs. Many people and companies promised to donate money. The organization wanted to share its research and patents openly. This was to help other researchers.
OpenAI first operated from Greg Brockman's living room. Later, it moved to the Pioneer Building in San Francisco. OpenAI offered good salaries to attract top AI researchers. These researchers were drawn to OpenAI's goal of developing AI safely.
In April 2016, OpenAI launched "OpenAI Gym." This was a platform for reinforcement learning research. Nvidia gave OpenAI a powerful computer in August 2016. This computer helped train large AI models faster. In December 2016, OpenAI released "Universe." This software platform helped train AI's general intelligence. It used games, websites, and other applications.
Training AI models required a lot of computing power. For example, in 2018, training OpenAI's Dota 2 bots needed many computers for weeks.
In 2018, Elon Musk left OpenAI's board. He mentioned a possible conflict of interest with his company, Tesla. Sam Altman said Musk wanted to take over OpenAI, but the board disagreed.
In February 2019, OpenAI announced GPT-2. This AI model could create text that sounded very human-like.
Becoming a "Capped" For-Profit Company (2019)
In 2019, OpenAI changed its structure. It went from a non-profit to a "capped" for-profit company. This means profits are limited to a certain amount. This change allowed OpenAI to get investments from companies. It also let them offer company shares to employees. This helps attract top researchers who might otherwise work for big tech companies.
After this change, Microsoft invested $1 billion in OpenAI. Since then, OpenAI's systems have run on Microsoft's Azure supercomputing platform. OpenAI Global, LLC also planned to offer its technology for commercial use. Sam Altman said that even a billion dollars might not be enough. He believed the lab might need much more money to achieve artificial general intelligence.
Some people were unsure about this change. They wondered if a for-profit company could still focus on the public good. However, the non-profit OpenAI, Inc. still controls the for-profit part. Its main goal remains to benefit humanity.
ChatGPT, DALL-E, and Microsoft Partnership (2020–2023)
In 2020, OpenAI introduced GPT-3. This is a large language model trained on huge amounts of internet data. GPT-3 can answer questions, translate languages, and write creative text. OpenAI also launched an API (Application Programming Interface) for its first commercial product.
In 2021, OpenAI released DALL-E. This AI model creates digital images from text descriptions. It uses a version of the GPT-3 technology.
In December 2022, OpenAI launched a free preview of ChatGPT. This new AI chatbot quickly became very popular. It gained over a million users in just five days. By January 2023, ChatGPT was the fastest-growing consumer software ever. It reached 100 million users in two months.
In January 2023, Microsoft announced a new $10 billion investment in OpenAI. This investment was partly for using Microsoft's cloud-computing service, Azure. This deal was part of Microsoft's plan to add ChatGPT to its Bing search engine. Other companies, like Google, also started developing similar AI tools.
On March 14, 2023, OpenAI released GPT-4. This model was available through an API and for ChatGPT Plus users. GPT-4 could accept both text and image inputs.
In May 2023, Sam Altman and other leaders shared ideas for managing "superintelligence." They suggested that superintelligence could arrive within 10 years. They proposed creating an international group to oversee powerful AI systems. They also called for more research into AI safety.
In July 2023, OpenAI started the "superalignment project." This project aims to find ways to make future superintelligent AIs follow human values. In August 2023, OpenAI bought Global Illumination, a startup that uses AI for digital tools.
In October 2023, OpenAI announced a partnership with the Emirati AI firm G42. This would allow G42 to use OpenAI's technology. On November 6, 2023, OpenAI launched GPTs. This feature lets people create their own versions of ChatGPT for specific tasks.
Recent Developments (2024)
In January 2024, OpenAI partnered with Arizona State University. This was OpenAI's first partnership with a school. It gave the university full access to ChatGPT Enterprise.
On February 15, 2024, OpenAI announced Sora. This AI model creates videos from text descriptions. OpenAI plans to release it to the public later.
On February 29, 2024, Elon Musk filed a lawsuit against OpenAI and Sam Altman. He claimed they were focusing on profits instead of their original mission to benefit humanity. OpenAI responded that Musk understood their mission did not mean making all AI open-source. They denied being controlled by Microsoft. In June, Musk withdrew the lawsuit, but later reopened it.
In May 2024, some key leaders and AI safety researchers left OpenAI. This led to the superalignment group being closed down.
On May 19, 2024, Reddit and OpenAI announced a partnership. Reddit's content will be used to improve OpenAI products like ChatGPT. This helps OpenAI access real-time information. Reddit also plans to create new AI features for its users.
OpenAI also made deals with news companies like News Corp, Vox Media, and The Atlantic. These deals allow OpenAI to use their news content to make AI models more accurate. However, other publications, like The New York Times, have filed lawsuits against OpenAI and Microsoft. These lawsuits are about using their content to train AI models without permission.
A group of current and former OpenAI employees raised concerns. They said the company was moving too fast with AI development. They called for more openness and rules for AI.
On June 10, 2024, OpenAI partnered with Apple Inc.. This will bring ChatGPT features to Apple Intelligence and iPhone. On June 13, 2024, Paul Nakasone, former head of the NSA, joined OpenAI's board.
In July 2024, OpenAI began working on a project to improve AI's reasoning. This project, called o1, was released on September 12. It helps AI plan and do "deep research."
In September 2024, Mira Murati, OpenAI's Chief Technology Officer, left the company. In October 2024, OpenAI raised $6.6 billion from investors. This valued the company at $157 billion. Major investors included Microsoft, Nvidia, and SoftBank.
In November 2024, Greg Brockman rejoined OpenAI as president. OpenAI also acquired the domain Chat.com, redirecting it to ChatGPT's main site. In December 2024, OpenAI launched several new features. These included Sora for video creation and the o1 model for better reasoning. They also introduced ChatGPT Pro, a paid service with advanced features.
Plans for 2025
In January 2025, a model called "DeepSeek-R1" was released by DeepSeek. It performed as well as OpenAI's o1 model but cost much less to train. This caused a big drop in Nvidia's market value.
On January 21, 2025, OpenAI, Oracle, SoftBank, and MGX announced "The Stargate Project." This is a huge plan to build an AI infrastructure system with the U.S. government. It is estimated to cost $500 billion.
On January 24, OpenAI launched Operator. This AI tool helps users access websites and complete tasks. It was first available to Pro users in the United States.
In February 2025, OpenAI changed its brand with a new look. They also started working with Broadcom to design a special AI chip. This chip will help reduce their reliance on Nvidia GPUs. Sam Altman announced that GPT-4.5 would be the last model without full "chain-of-thought" reasoning. He also said that GPT-5 might combine the O-Series and GPT-Series models.
How OpenAI is Managed
Key Leaders
- CEO and co-founder: Sam Altman, who used to lead Y Combinator.
- President and co-founder: Greg Brockman, who was a CTO at Stripe.
- Chief Scientist Officer: Jakub Pachocki, former Director of Research at OpenAI.
- Chief Operating Officer: Brad Lightcap.
- Chief Financial Officer: Sarah Friar, who was CEO of Nextdoor.
- Chief Product Officer: Kevin Weil.
- Chief Compliance Officer: Scott Schools.
Board of Directors for the OpenAI Non-Profit
- Bret Taylor (chairman), former chairman of Twitter's board.
- Sam Altman
- Lawrence Summers, former U.S. Secretary of the Treasury.
- Adam D'Angelo, co-founder and CEO of Quora.
- Sue Desmond-Hellmann, former CEO of the Bill & Melinda Gates Foundation.
- Nicole Seligman, an attorney.
- Fidji Simo, CEO of Instacart.
- Paul Nakasone, former Director of the National Security Agency.
- Zico Kolter, a computer scientist.
- Adebayo Ogunlesi, managing partner at Global Infrastructure Partners.
Main Individual Investors
- Reid Hoffman, co-founder of LinkedIn.
- Peter Thiel, co-founder of PayPal.
- Jessica Livingston, a founding partner of Y Combinator.
- Elon Musk, co-founder.
Why OpenAI Started
Some scientists, like Stephen Hawking, have worried about advanced AI. They fear that if AI can improve itself very quickly, it could lead to "intelligence explosion" and even human extinction. Elon Musk calls AI humanity's "biggest existential threat."
Musk and Altman started OpenAI partly because they were concerned about AI safety. They believe AI could greatly help society. But they also know it could cause harm if not built or used correctly. They say it is hard to predict when human-level AI will arrive. OpenAI believes AI "should be an extension of individual human wills." It should be shared widely and fairly.
Vishal Sikka, former CEO of Infosys, supported OpenAI. He liked its goal to "produce results generally in the greater interest of humanity." OpenAI's mission to make AI safe and beneficial attracted many researchers.
OpenAI's Approach
Elon Musk and Sam Altman had a unique idea. They thought that giving many people access to AI could reduce its potential harm. Musk said, "If everyone has AI powers, then there's not any one person or a small set of individuals who can have AI superpower."
This idea is debated by some who worry about AI risks. Philosopher Nick Bostrom said, "If you have a button that could do bad things to the world, you don't want to give it to everyone."
OpenAI's decision in 2019 not to immediately release GPT-2 caused discussion. They wanted to be careful because of possible misuse. Some experts doubted GPT-2 was truly dangerous. Others argued that sharing the research was important to understand it and create ways to counter misuse.
More recently, in 2022, OpenAI shared its plan for "AI alignment." This is about making sure AI, especially AGI, follows human values. They believe that if AGI is not aligned, it could be very risky for humanity. They plan to use human feedback to train AI systems. They also want to use AI to help with alignment research.
In 2024, after Sam Altman's temporary removal and return, some employees left OpenAI. The company also planned to restructure as a for-profit company. This could give Altman a stake in the company.
Stance on China
In February 2025, OpenAI CEO Sam Altman said the company wants to work with China. This is despite rules from the U.S. government. This interest comes as Chinese AI companies like DeepSeek are growing fast. DeepSeek has released powerful and cost-effective AI models.
The rise of DeepSeek has made many Chinese tech firms use an open-source approach. This increases competition with OpenAI. Altman knows that U.S. government approval for AI cooperation with China is uncertain. But he stressed the importance of talking between tech leaders in both countries.
Products and Applications
Learning Through Practice (Reinforcement Learning)
In its early years, OpenAI focused on reinforcement learning (RL). This is a type of AI training where the AI learns by trying things and getting rewards or punishments. OpenAI was a key competitor to Google DeepMind.
Gym
Launched in 2016, Gym is a free software library for Python. It helps create and test reinforcement learning algorithms. It makes it easier to share and repeat AI research. In 2022, new updates for Gym moved to a library called Gymnasium.
Gym Retro
Released in 2018, Gym Retro is a platform for RL research using video games. It helps AI learn to play games and apply what it learns to new situations. This is called "generalization."
RoboSumo
RoboSumo, released in 2017, is a virtual world. Here, robot agents learn to move and push other robots out of a ring. They start without knowing how to walk. Through this competition, they learn to adapt. For example, if placed in a windy environment, they learn to balance. This shows they learned to balance in a general way.
OpenAI Five
OpenAI Five is a team of five AI bots that play the video game Dota 2. They learned to play against human players at a high level. They learned only by practicing through trial and error.
In 2017, one bot played against a professional player and won. The bot learned by playing against itself for two weeks. This showed how AI could handle complex tasks. The bots learned by playing thousands of games against themselves.
By June 2018, the bots could play as a full team of five. They beat amateur and semi-professional teams. In April 2019, OpenAI Five defeated OG, the world champions of Dota 2, in a match. The bots played over 42,000 games in an online competition. They won 99.4% of them.
Dactyl
Dactyl, developed in 2018, uses machine learning to control a robot hand. This hand can move physical objects. It learns by practicing in a virtual world. OpenAI used "domain randomization" to help the robot learn. This means the robot practices in many different virtual situations. This helps it adapt to real-world changes.
In 2019, OpenAI showed that Dactyl could solve a Rubik's Cube. The robot solved the puzzle 60% of the time. They improved Dactyl by using "Automatic Domain Randomization" (ADR). ADR creates harder environments for the robot to learn in.
API (Application Programming Interface)
In June 2020, OpenAI launched a multi-purpose API. This API lets developers use OpenAI's AI models for "any English language AI task."
Creating Text (Text Generation)
OpenAI has made generative pretrained transformers (GPT) very popular.
OpenAI's First GPT Model ("GPT-1")
The first paper on GPT models was published in June 2018. It showed how a language model could learn about the world. It did this by reading a lot of text.
GPT-2
Generative Pre-trained Transformer 2 ("GPT-2") was announced in February 2019. It was an improved version of GPT-1. Only limited versions were released at first. This was due to worries about it being used to write fake news. Some experts were unsure if GPT-2 was truly dangerous.
In response, the Allen Institute for Artificial Intelligence created a tool to find "neural fake news." Other researchers warned that AI could fill the internet with realistic-sounding fake text. In November 2019, OpenAI released the full GPT-2 model.
GPT-2 was very good at learning general tasks. It achieved high accuracy on many tasks without extra training. It was trained on a large dataset of text from the internet.
GPT-3
Generative Pre-trained Transformer 3 (GPT-3) was first described in May 2020. It is much larger than GPT-2. The full GPT-3 model has 175 billion "parameters." These are the parts of the model that learn from data.
OpenAI said GPT-3 could do "meta-learning" tasks. This means it could learn from just a few examples. GPT-3 showed big improvements over GPT-2. OpenAI warned that even with such large models, there might be limits to what they can do. Training GPT-3 needed a huge amount of computing power. The full GPT-3 model was not immediately released to the public. OpenAI planned to offer access through a paid online service.
On September 23, 2020, Microsoft got an exclusive license to use GPT-3.
Codex
Codex was announced in mid-2021. It is based on GPT-3 but also trained on computer code. It powers GitHub Copilot, a tool that helps programmers write code. Codex can create working code in many programming languages.
There were some issues with Codex, including glitches and security problems. GitHub Copilot was also accused of using copyrighted code without giving credit. OpenAI stopped supporting the Codex API on March 23, 2023.
GPT-4
On March 14, 2023, OpenAI released Generative Pre-trained Transformer 4 (GPT-4). This model can understand both text and images. OpenAI said GPT-4 passed a law school exam with a score in the top 10%. (GPT-3.5 scored in the bottom 10%.) GPT-4 can also read, analyze, or create up to 25,000 words of text. It can also write code in all major programming languages.
People noticed that ChatGPT using GPT-4 was much better than the previous version. However, GPT-4 still had some of the same issues as older versions. OpenAI has not shared many technical details about GPT-4.
GPT-4o
On May 13, 2024, OpenAI announced and released GPT-4o. This model can process and create text, images, and audio. GPT-4o performed very well in tests for voice, multiple languages, and vision. It set new records for understanding spoken language and translation.
On July 18, 2024, OpenAI released GPT-4o mini. This is a smaller version of GPT-4o. It replaced GPT-3.5 Turbo on the ChatGPT interface. It is also cheaper to use through its API. OpenAI expects it to be useful for businesses and developers.
o1
On September 12, 2024, OpenAI released the o1-preview and o1-mini models. These models are designed to "think" longer about their answers. This leads to more accurate results. They are especially good at science, coding, and reasoning tasks. These models were made available to ChatGPT Plus and Team members. In December 2024, o1-preview was replaced by o1.
o3
On December 20, 2024, OpenAI showed off o3. This is the next version of their o1 reasoning model. They also showed o3-mini, which is smaller and faster. As of December 2024, these models are not yet available to the public. OpenAI is testing them. The model is called o3 instead of o2 to avoid confusion with a phone company.
Deep Research
Deep research is an agent developed by OpenAI. It was shown on February 2, 2025. It uses OpenAI's o3 model to browse the internet, analyze data, and create detailed reports. It can do this in 5 to 30 minutes. With internet and coding tools, it achieved 26.6% accuracy on a difficult test called HLE (Humanity's Last Exam).
Image Understanding (Image Classification)
CLIP
CLIP (Contrastive Language–Image Pre-training) was revealed in 2021. It is a model trained to understand how text and images are related in meaning. It can be used to sort images into categories.
Creating Images from Text (Text-to-Image)
DALL-E
DALL-E was revealed in 2021. It is an AI model that creates images from text descriptions. DALL-E uses a version of GPT-3 to understand text like "a green leather purse shaped like a pentagon." Then, it creates matching images. It can make pictures of real things or things that don't exist, like "a cube with the texture of a porcupine."
DALL-E 2
In April 2022, OpenAI announced DALL-E 2. This is an updated version that creates more realistic images. In December 2022, OpenAI released software for Point-E. This is a new system that can turn text into a 3D model.
DALL-E 3
In September 2023, OpenAI announced DALL-E 3. This is a more powerful model. It can create images from complex descriptions without needing special instructions. It also handles details like hands and text better. It was released to the public as a ChatGPT Plus feature in October.
Creating Videos from Text (Text-to-Video)
Sora
Sora is a text-to-video model. It can create videos from short text descriptions. It can also extend existing videos. It can make videos up to 1920x1080 resolution.
Sora's name comes from the Japanese word for "sky." This means its "limitless creative potential." Sora uses technology similar to DALL-E 3. OpenAI trained Sora using publicly available videos and some licensed videos.
On February 15, 2024, OpenAI showed some videos made by Sora. They said it could make videos up to one minute long. They also shared a report about how it works. They noted some weaknesses, like struggling to simulate complex physics. Some experts were impressed, but others wondered if the videos were the best examples.
Even with some doubts, people in the entertainment industry are very interested in Sora. Filmmaker Tyler Perry was amazed by it. He said it could change how stories are told and content is made. He even paused plans to expand his movie studio because of Sora's potential.
Turning Speech into Text (Speech-to-text)
Whisper
Whisper was released in 2022. It is a general-purpose model for recognizing speech. It was trained on a large amount of different audio. It can recognize speech in many languages. It can also translate speech and identify languages.
Creating Music (Music Generation)
MuseNet
MuseNet was released in 2019. It is a deep neural network that predicts musical notes in MIDI files. It can create songs with 10 instruments in 15 styles. However, songs made by MuseNet sometimes become chaotic as they play longer.
Jukebox
Jukebox was released in 2020. It is a free algorithm that creates music with vocals. It was trained on 1.2 million music samples. Users can give it a music style, artist, and some lyrics. Then, it creates song samples. OpenAI said the songs have good musical flow but lack repeating parts like choruses. They noted a big difference between Jukebox and human-made music.
User Interfaces
Debate Game
In 2018, OpenAI launched the Debate Game. This game teaches machines to debate simple problems in front of a human judge. The goal is to see if this can help understand how AI makes decisions. It also helps develop explainable AI.
Microscope
Microscope was released in 2020. It is a collection of pictures that show the inner workings of neural networks. It helps people easily study how these AI models work. It includes models like AlexNet and VGG-19.
ChatGPT
.svg){kind=logo}
Launched in November 2022, ChatGPT is an AI tool built on GPT-3. It has a chat interface. Users can ask questions in everyday language, and the system answers in seconds. ChatGPT reached 1 million users in just 5 days.
As of 2023, ChatGPT Plus is a version that uses GPT-4. It costs $20 per month. The original version uses GPT-3.5. OpenAI also offers GPT-4 through an API for developers. This costs a fee based on how much text is processed.
In May 2023, OpenAI launched a ChatGPT app for iOS and later for Android. The app saves chat history and allows voice input. In September 2023, OpenAI announced that ChatGPT "can now see, hear, and speak." ChatGPT Plus users can upload images. Mobile app users can talk to the chatbot.
In October 2023, DALL-E 3 was added to ChatGPT Plus and Enterprise. ChatGPT helps write prompts for DALL-E based on user conversations.
There are growing concerns about how much energy generative AI, like ChatGPT, uses. In September 2024, Microsoft made a deal to reopen a nuclear power plant. This is to supply power to its AI data centers.
In December 2024, OpenAI launched a new feature. It allows users to call ChatGPT for up to 15 minutes per month for free.
SearchGPT
SearchGPT is a test search engine developed by OpenAI. It was shown on July 25, 2024, to a small group of users. It combines regular search features with AI capabilities.
Stargate and Other Supercomputers
Stargate is a huge artificial intelligence supercomputer being developed by Microsoft and OpenAI. They are working with Oracle, SoftBank, and MGX. Stargate is part of a larger data center project. This project could cost up to $100 billion from Microsoft.
Stargate is planned as the fifth and final phase of a series of AI construction projects. It is expected to be finished around 2028. The supercomputer will use millions of special server chips. Its data center will be built in the U.S. and will need a lot of power. It might use nuclear energy. The name "Stargate" is a nod to the 1994 sci-fi movie Stargate.
See also
In Spanish: OpenAI para niños
- Anthropic
- Center for AI Safety
- Future of Humanity Institute
- Future of Life Institute
- Google DeepMind
- Machine Intelligence Research Institute
- Microsoft
- Bing
- The New York Times
Images for kids
-
OpenAI logo
-
OpenAI's "Blossom" symbol
