It feels like Christmas for those of us obsessed with AI. On Monday, November 6th, OpenAI hosted its first-ever DevDay. The event kicked off with a warm welcome and an exciting atmosphere in San Francisco. The goal? To celebrate the progress that OpenAI has achieved over the past year while showcasing new technology that would launch the creativity and efficiency of its customers. 

Let’s recap the highlights of the presentation. Since AIPRM and ChatGPT currently have a symbiotic relationship, we wanted to cover the significant announcements and insights shared by OpenAI’s leadership. Despite the upgrades, AIPRM is still the ChatGPT cheat code.

Looking Back: OpenAI’s Journey

The presentation began with a quick look back at OpenAI’s recent achievements. CEO of OpenAI, Sam Altman shared the story of a tool that went viral. 

About a year ago, on November 30th to be precise, OpenAI introduced ChatGPT in a low-key research preview. This initial launch set the stage for a series of advancements and mainstream commercialization of AI technology. 

In March, GPT-4 was unveiled, a powerful upgrade to the previous 3.5 model with increased speed and reasoning capabilities. We also got access to 3rd Party Plugins that added the ability to connect to the internet.

In May, we got an iOS App for ChatGPT.

In August, OpenAI launched ChatGPT Enterprise, offering enhanced security, privacy, faster GPT-4 access, and more.

In September, OpenAI expanded ChatGPT’s functionality, enabling it to see, hear, and speak. 

In October, DALL-E 3, the world’s most advanced image model, was also introduced as an integration within ChatGPT.

OpenAI’s commitment to developers was evident, with approximately 2 million developers using the API for a wide range of applications. Notably, over 92% of Fortune 500 companies have embraced OpenAI’s products. 

The presentation proudly mentioned that ChatGPT now boasts around 100 million weekly active users, a remarkable achievement driven entirely by word-of-mouth recommendations.

User Stories: The Heart of OpenAI’s Success

While numbers are impressive, the heart of OpenAI’s success lies in how people use their products. To illustrate this, a touching video showcased the diverse ways users have harnessed AI technology for personal and professional purposes. Users expressed how AI has empowered them to communicate, create, learn, and even make daily tasks more manageable. The video highlighted the human element in AI and how OpenAI’s technology has had a profound impact on users’ lives.

New Developments: GPT-4 Turbo

OpenAI transitioned to the main focus of the presentation—new developments. Sam acknowledged the valuable feedback received from developers worldwide and announced a new model: GPT-4 Turbo. This model addresses various developer requests and comes with several notable enhancements.

Extended Context Length: GPT-4 Turbo supports an impressive context window of up to 128,000 tokens, accommodating even the most extensive tasks, with greater accuracy over long contexts.

Enhanced Control: Developers will appreciate the introduction of JSON mode, which ensures that the model’s responses are in valid JSON format. The model is also more adept at handling function calls and following instructions. A new feature, reproducible outputs, allows users to achieve consistent model behavior.

Better World Knowledge: OpenAI has launched retrieval in the platform, enabling the incorporation of external knowledge from documents and databases. GPT-4 Turbo’s knowledge is up to date until April 2023, and OpenAI commits to maintaining this knowledge over time.

New Modalities: OpenAI introduces DALL-E 3, GPT-4 Turbo with Vision, and a new text-to-speech model in the API. These additions expand the possibilities for creating images, designs, and natural-sounding audio.

Customization: Developers can now fine-tune the 16K version of the model, and OpenAI offers an experimental access program for GPT-4 fine-tuning. The Custom Models program allows close collaboration with companies to create customized models tailored to specific domains.

Higher Rate Limits: OpenAI is doubling the tokens per minute for established GPT-4 customers, making it easier for developers to accomplish more. They are also introducing Copyright Shield, which provides legal protection to customers against copyright infringement claims.

Pricing Update: More Affordable GPT-4 Turbo

One of the most significant developer requests was related to pricing. In response to this feedback, OpenAI has made GPT-4 Turbo considerably more affordable than GPT-4. The pricing has been reduced by a factor of 3x for prompt tokens and 2x for completion tokens, making GPT-4 Turbo a cost-effective choice for developers.

OpenAI is showcasing commitment to its users in these pricing changes, with a promise to prioritize speed improvements in the near future.

GPTs: A New Era of Customizable AI

OpenAI recognizes that AI is evolving, and people want smarter, more personalized, and customizable AI agents. The idea of agents, that can handle complex tasks and adapt to user needs, is the future OpenAI envisions. To make this a reality, they have introduced GPTs – customized versions of ChatGPT designed for specific purposes.

Use Cases for GPTs

GPTs are designed to be versatile and can be tailored to a wide range of applications. Let’s explore a few real-world examples:

Code.org’s Lesson Planner GPT

Code.org, an organization dedicated to expanding computer science education in schools, created a GPT called Lesson Planner. This GPT helps teachers provide engaging lessons to middle school students. Teachers can ask for explanations on complex topics using creative and understandable language, making learning more enjoyable for students.

Canva’s Design Assistant GPT

Canva, a popular graphic design platform, introduced a GPT that allows users to create designs by describing their vision in natural language. Users can simply provide details, such as “Make a poster for the dev day reception this evening,” and the GPT generates design options, leveraging Canva’s APIs.

Zapier’s Integration Helper GPT

Zapier, a platform for connecting various applications, built a GPT to assist users in performing actions across 6,000 applications. Users can instruct the GPT to carry out specific tasks, such as checking their daily schedule or sending messages, seamlessly integrating various tools and automating tasks.

How to Build Your Own GPT

OpenAI has made it easy for developers to create their own GPTs, tailored for specific purposes. The GPT Builder provides a user-friendly interface to craft your agent. Here’s a quick overview of the process:

Define Your GPT’s Purpose: Start by describing the intended use of your GPT and give it a suitable name.

Customize Instructions: Specify the instructions, expanded knowledge, and actions you want your GPT to perform.

Configure Capabilities: Enable features that align with your GPT’s purpose, such as custom actions or specific questions.

Interaction Style: Describe the desired interaction style or user experience.

Test and Refine: Try out your GPT to ensure it functions as intended.

Publish and Share: You can choose to keep your GPT private, share it with the public, or limit it to your company.

GPT Store and Revenue Sharing:

OpenAI plans to launch the GPT Store later this month. It will allow developers to list their GPTs and will feature the most useful and popular agents. Revenue sharing is also part of the plan, with OpenAI offering compensation to those who create valuable GPTs.

Extending GPTs to the API

The most exciting part for developers is that OpenAI is bringing the GPT concept to the API. Many have already built agent-like experiences on the API, creating customized AI interactions that can perform specific tasks and respond to user requests.

OpenAI’s introduction of GPTs opens up a world of possibilities for developers, allowing them to create customized AI agents for various applications. These agents can provide personalized, interactive experiences and automate tasks, making AI more useful and accessible to everyone. With GPTs and the upcoming GPT Store, OpenAI appears to be fostering a vibrant ecosystem of AI development and innovation.

The New Assistance API: Making AI Development Easier

As mentioned by the OpenAI team, building custom AI assistants could be a complex and time-consuming process, often involving teams of engineers working for months. However, OpenAI is now introducing a new solution – the Assistance API. This API is designed to simplify the creation of AI-powered assistants for a wide range of applications.

Key Features of the Assistance API:

Persistent Threads: Developers can create threads for users to streamline conversations without the need to manage extensive conversation histories manually.

Built-in Retrieval: The API includes a retrieval system that enables the assistant to access knowledge beyond immediate user messages. This enhances the AI’s ability to provide accurate and contextually relevant responses.

Code Interpreter: The code interpreter allows the AI to write and execute code on the fly, making it a powerful tool for developers to automate complex tasks and calculations.

Function Calling: Building on earlier improvements, the updated Function Calling feature guarantees JSON output with no added latency, and developers can now invoke multiple functions at once.

Assistance API Live Demo

Roman, OpenAI’s Head of Developer Experience, demonstrated how easy it is to create an AI assistant using the new Assistance API. In his example, he was building an AI-powered travel app called “Wanderlust.”

Creating an Assistant: The process begins by giving the assistant a name, specifying initial instructions, and selecting the model (in this case, GPT-4 Turbo). Developers can also choose from tools like the Code Interpreter.

Threads and Messages: Developers can create threads for each user and add messages to these threads as users interact with the assistant.

Function Calling in Action: Roman demonstrated how the assistant handled complex user requests, such as calculating travel expenses, and using Function Calling to generate code and provide answers.

Retrieval and Data Analysis (formerly Code Interpreter): Roman showcased how the assistant could extract information from uploaded files (e.g., flight tickets, AirBnB bookings), simplifying the integration of external knowledge into the conversation. The Data Analysis automatically handled calculations.

Voice Integration: Roman took it up a notch by showing how a voice-based assistant could interact with users. He used Whisper to convert voice inputs into text, and the assistant could perform tasks based on voice commands.

Internet Interaction: The assistant took a real-time action, selecting five Dev Day attendees and awarding them with $500 in OpenAI credits, demonstrating the power of function calling combined with the new modalities.

Out of sheer generosity, the assistant even changed its mind, distributing credits of equal value to all Dev Day attendees.

The OpenAI team emphasized the importance of gradual, iterative deployment, allowing developers to build and experiment with AI agents to understand their potential. OpenAI encourages users to provide feedback to drive further improvements.

The Assistance API is a groundbreaking tool that simplifies the development of AI assistants for various applications. With features like Function Calling, Voice Integration, Retrieval, and Code Interpreter, developers can unlock the full potential of AI to create innovative and customized experiences. 

The future of AI agents is taking shape, and OpenAI is at the forefront, driven by the belief that AI will empower individuals and transform the world in unprecedented ways.

Stay tuned for more updates as OpenAI continues to push the boundaries of AI technology and brings more advanced capabilities to developers. The possibilities are limitless, and the future looks incredibly promising.