Whyle we’ve got the announcement about GPT-4, OpenAI has provided the public access to the GPT-3 Natural Language Processing API (Application Programming Interface). This opens new opportunities for experimens with intelligent systems that can mimic human capabilities, such as writing poetry, machine code, or answering questions. If you need you own AI programming solution, I wish to recommend my partners, who’ve designed the tailor-made AI-based system for our product.
What is GPT-3
GPT-3 (Generative Pre-trained Transformer 3) is a universal natural language processing algorithm, the world’s largest artificial intelligence model. It allows the computer to understand the meaning of what is written by a person to perform specific actions.
GPT-3 was trained on 500 billion words and includes 175 billion parameters that artificial intelligence seeks to optimize during training. The model is capable of generating “eerily human text” in response to queries and parsing strings and drawing conclusions from unstructured data presented in writing.
The developers of OpenAI claim that the algorithms can be used to solve “any problem in English”. Based on GPT-3, systems have already been developed for translating from natural language into program code and automatic code completion, retelling the contents of books, and a video game that has become a business consultant.
Why an open API is important
Previously, developers had to sign up on a waiting list to get experimental access to the GPT-3 API. The solution creators were afraid that the neural network would facilitated for something like generating fake news.
Now, OpenAI has opened up full access to the GPT-3 API, deciding that the software is “safe enough to be effectively controlled.” An open API will allow developers to use GPT-3 to solve problems in various areas.
Restrictions on the use of the GPT-3 API remained only for industrial development. In addition, the developers have banned using the neural network to create content that can provoke outbreaks of hatred, influence the political process, or pose a danger to minors.
The GPT-3 open API is an opportunity to evaluate all the risks in practice, said Peter Welinder, vice president of products and partnerships at OpenAI. As long as the stakes are not too high, developers will be able to quickly identify possible problems and create solutions to improve the security level of the smart model.
Researchers from OpenAI decided to make GPT-3 more “transparent” and taught it to generate answers and provide links to sources. They created an interactive environment for GPT-3 – a text browser that works through the Bing search engine API. The environment shows the results for a search query with headings and a paragraph of text and also allows you to open the desired link, scroll through the page and perform other basic actions.
First, GPT-3 receives a question from the user and the state of the browser that asked the question to the search engine. Then the model starts surfing the web: opens links, read the text, and, if she sees the information she needs, saves it along with the title and address of the page. After each action (click, scroll, etc.), the model receives a new browser state. This continues until the model itself decides to end the search or the maximum number of actions or collected data is reached. In the last step, the model receives a query that includes the original question and collected information from the Internet. It generates a single final answer with links to specific statements. Discover more AI-related hot topics here.