The latest version of the OpenAI language model, GPT-4, has a new and impressive ability: it can use tools. This means that it can interact with external applications and services. OpenAI calls the new feature “function calling” and it is built directly into the GPT-4 API. The update follows the company's recent decision to make the GPT-4 API more affordable.
For example, GPT-4 can answer questions like “How far is it from New York to Paris?” or “Who is the president of Brazil?” by using the appropriate tool to find the answer. It can also generate content that incorporates data from these tools, such as maps, charts, or tables.
This is a big deal because it shows that GPT-4 is not just a passive consumer of text, but an active agent that can manipulate and create information. It also opens up new possibilities for what GPT-4 can do and how it can be used.
One possibility is that GPT-4 can become a powerful assistant that can help users with various tasks, such as planning trips, researching topics, or writing reports. Another possibility is that GPT-4 can become a creative partner that can generate novel and interesting content, such as stories, poems, or songs.
Of course, there are also challenges and risks associated with this new ability. For one thing, GPT-4 still has limitations in its understanding and reasoning, and may produce inaccurate or misleading results. For another thing, GPT-4 may access and use information that is sensitive or inappropriate, and may violate ethical or legal norms.
Therefore, it is important to monitor and regulate how GPT-4 uses tools, and to ensure that it respects the privacy and rights of users and data sources. It is also important to educate users about the capabilities and limitations of GPT-4, and to encourage them to use it responsibly and critically.
Expanding the Capabilities of GPT-4
GPT-4 is the latest version of OpenAI's generative pre-trained transformer (GPT) model, which uses deep learning to generate coherent and fluent text from a given prompt.
GPT-4 was released in February 2023, and it boasts 175 billion parameters, making it more than four times larger than GPT-3. It is a large language model (LLM) that Microsoft also uses as part of its Bing Chat AI search chatbot. However, OpenAI is not currently using GPT-4 in its ChatGPT bot.
The new update changes also allows GPT-4 to execute code using a special syntax. For example, if a user asks GPT-4 to calculate the square root of 25, GPT-4 can reply with =sqrt(25) and then display the result 5. Similarly, if a user asks GPT-4 to draw a circle, GPT-4 can reply with =draw_circle() and then generate an image of a circle.
OpenAI claims that GPT-4 is the most advanced chatbot in the world, and that it can handle complex and diverse conversations with ease. However, the organization also acknowledges that GPT-4 is not perfect, and that it still has limitations and biases that need to be addressed. Therefore, OpenAI is inviting developers and researchers to test and improve GPT-4 through its API service.