OpenAI is a research company that started in 2015 in Silicon Valley, the global hub for disruptive technologies. The company was a result of the group efforts and investment from Elon Musk; Sam Altman; Greg Brockman; Wojciech Zaremba; Ilya Sutskever, and John Schulman. Though Musk later left his board seat because of a possible conflict of interest related to Tesla’s AI research and other concerns.
OpenAI primarily aims to provide smart and safe artificial intelligence tools that empower humanity. Its main focus lies in the study and development of artificial general intelligence (AGI), a type of AI technology that has a high reasoning, that can learn from experience and is able to complete challenging tasks.
The company has already landed an impressive range of AI tools, many of which you may have already heard of. Some examples include the AI image generator DALL·E, and GitHub’s coding tool CoPilot. If neither of these ring a bell, then we’re sure our next example will! The famous ChatGPT. Before we get into details, we’ll take a trip back in time and dive into the company’s history so you can better understand where ChatGPT and the other AI tools come from.
The Beginning of Open AI
OpenAI originally began operating as a nonprofit organisation that sought free collaborations with other institutions and researchers to keep their discoveries open to the public eye. In 2016, it launched one of its very first set of tools for reinforcement learning (RL) called OpenAI Gym.
Reinforcement learning is a subset of machine learning specialised on decision making and motor control. So, OpenAI Gym looked specifically to train AI agents and improve their capacity to achieve high-complex goals. The company later introduced Universe, a software designed to evaluate and train AI’s general intelligence in video games, websites and other applications.
The Beginning of GPT
Between 2017 and 2019, OpenAI focused on carrying out further research and development. It was in 2018, where they introduced the concept of a Generative Pre-trained Transformer (GPT). In simple terms, a GPT is made up of machine learning models called neural networks that resemble the structure of the human brain and that are trained on large text-based datasets. In this way, a GPT’s functions can vary between generating, answering questions and other tasks. The team developed their first language model GPT-1, which was trained using a big text dataset composed of over 7,000 unpublished books titled BookCorpus. The model subsequently was adjusted into the more powerful GPT-2, which was trained on 8 million web pages and allowed a more accurate text prediction.
With the improvements done to this model, the team realised how big of an accomplishment yet risk was GPT-2 to the rest of the world. Therefore, they decided not to make it available to the public out of fear that they would misuse it for scams, fake news and other malpractices. This event along with the desire of obtaining financial sustainability and leveraging the commercial applications of the technology is what prompted the company in 2019 to shift apart from its original “nonprofit” structure and opted instead, for a “capped-profit” organisation. This hybrid form of for-profit and nonprofit welcomes investments from different sources, in order to assign their profits to talent and equipment. In fact, Microsoft happens to be one of OpenAI’s commercial partners and the way that both have benefitted from each other will be mentioned further in this article.
Introducing DALL-E, GPT-3 and ChatGPT
In 2021, OpenAI introduced its new tool DALL-E, which combines a similar architecture as GPT-2 with deep learning models to process natural language and ultimately, function. DALL-E blends in concepts, attributes and styles to create realistic images and art based on text descriptions. It’s no surprise they would name the model after the Spanish artist, Salvador Dalí since what it is capable of seems pretty surreal.
The following year was of many improvements and technological achievements for OpenAI. The company trained its previous two versions of GPT with a much larger text dataset and the outcome was a smarter, faster and more powerful GPT-3. This was possible thanks to Microsoft’s partnership that provided OpenAI with a supercomputer able to process terabytes of information.
Later in 2022, not only did the company release DALL-E 2, which gave back more realistic and accurate results with higher resolution, but it also presented the well-known ChatGPT. This language model chatbot was built on top of GPT-3, having the ability to understand context, generate more specific answers and constantly improve based on previous conversations.
The significant funding from Microsoft has strengthened the relationship with OpenAI through the years, but the results of its partnership were even more evident in the announcement made at the beginning of 2023. Microsoft notified that ChatGPT was to be applied in two of its star products, Microsoft Edge and its Bing search engine. The idea is to enhance search capabilities and provide a new chat experience. With Microsoft working closely with OpenAI we could be surprised with more AI developments in the near future.
DALL-E 2 and ChatGPT have been a breakthrough in AI research, and there’s no doubt that AI technology will continue to evolve and take a significant role in our lives.
GPT-4: the New AI Beast
With a more sophisticated architecture, increased training data and more complex algorithms, GPT-4 was launched in just 6 months after its previous version. Some of its distinguishing attributes are its high level of response accuracy and of compliance to OpenAI’s content policy. In fact, OpenAI states that GPT-4 is 82% less likely to create content that breaches the company’s content regulations.
Because it can process visual input, you’re now also able to include text and image-based prompts. Additionally, it removes one of ChatGPT's main restrictions and raises the token capacity to 8,192 tokens, or nearly 25,000 words.
The company also announced that GPT-4 is more innovative and synergistic than any of its predecessors, being able to carry out creative and technical writing skills including song composition and screenwriting. All of this takes the user’s writing style into account.
A variety of apps, including Duolingo, Stripe, and Taskade, are already powered by GPT-4. However, GPT-4 is available to users by subscribing to ChatGPT Plus.
The Benefits of OpenAI
OpenAI has been transforming the way we engage with technology and it’s very likely to continue doing so in the next few years. With the help of OpenAI's research, AI systems can be refined, enabling more intuitive and natural interactions between humans and machines. More specifically, the emergence of stronger neural networks could be the key to tackling complex issues that would be challenging for people to deal with on their own.
OpenAI counts with equipment, financial support, and some of the best talent. All of which have helped OpenAI position itself as the leader in AI. The company can leverage its authority to create partnerships across different industries and thus, cross-pollinating ideas. In 2022, LinkedIn’s co-founder and partner at Greylock, Reid Hoffman, began working with OpenAI and AdeptAI laboratories with the purpose of creating AI technologies that "empower humans" rather than replace us in the workforce.
Entering new markets such as healthcare, transportation, and finance, would help OpenAI unlock opportunities to develop new technologies and access a sizable amount of real-world customer data to train its AI models. X-ray and CT (computer tomography) scan reviews, patient record analysis, public transportation management, and even agricultural yield optimisation are all currently being done by AI systems.
ChatGPT and other similar tools can benefit other fields including education, marketing, financial services, and customer care. But it’s important to bear in mind that with such benefits, a few risks may be presented as well.
The future of AI will depend on companies like OpenAI, as they’ve been shaping the panorama and defining what "safe" and "beneficial" use entails.
The Challenges of OpenAI
Despite the progress made by OpenAI, the research on AI is still in its early stages. There are numerous challenges that will need to be addressed before AI research becomes as “open” as the company initially envisioned.
Assuring the security and dependability of OpenAI's AI systems is one of the main concerns. OpenAI must make sure that its AI systems are not biased towards any particular demographic or used with bad intentions. Moreover, OpenAI must make sure that its AI systems are impervious to abuse or manipulation.
While it’s true that OpenAI drifted apart from its original objective of making AI research and tools available to everyone, it’s also true that the company has disclosed the issues of ethical AI development.
However, it’s been kind of ironic that the company claims to worry about AI ethics when DALL-E 2 was being unethical by nature. Indeed, some artists have found out that their work was being plagiarised by the tool. To make it worse, some even noticed evidence of their blurred signatures in the images, which would explain why many showed their discontent with the way OpenAI had collected customer data at the time.
ChatGPT has also caused some distress in the academic area, as many researchers and teachers believe that the performance and credibility of students may be compromised by AI tools. Many have been caught cheating on school assignments and examinations. With this example, it is even more clear how both the developers of AI systems and their users are accountable for the ethical and safe use of those technologies.
OpenAI has been subtly drifting away from its initial objective of open, accessible, and safe AI development ever after the GPT-2 launch was postponed. The team has also received high criticism for keeping its financial operations hidden from the public. Because creating AI systems is absurdly expensive, OpenAI must balance its purpose with maintaining its financial stability.
Now that we’ve covered what OpenAI is and does, as well as some of the benefits and challenges it’s presented to the world, you can expand your AI knowledge by checking out our AI Hub!