In this article, we will explore the incredible developments and improvements made by OpenAI. Numerous breakthroughs have been made in AI capabilities, leading to millions of developers utilizing AI for multifarious use cases.
The focus of this article is to discuss these substantial developments in AI and share stories of how users worldwide have benefitted from the technology. Moreover, we will take a closer look at the exciting advancements coming soon in the AI landscape that promise even more incredible opportunities for users in the future.
Journey So Far
- ChatGPT Release: About a year ago, ChatGPT was initially introduced as a “low-key research preview,” garnering positive feedback.
- GPT-4 Launch: In March 2023, the competent GPT-4 model was launched, which remains the most powerful model available.
- Voice and Vision Capabilities: Over the past few months, ChatGPT has gained the ability to perceive, understand, and interpret visual and auditory information.
- DALL-E 3 Launch: The world’s most advanced image model, DALL-E 3, was recently introduced and can be utilized within ChatGPT.
- ChatGPT Enterprise Launch: Offering enterprise-grade security and privacy, faster GPT-4 access, and longer context windows, ChatGPT Enterprise was launched for business clients.
📈 Stats and Facts: Today, about 2 million developers are using the API to create a diverse range of applications, while more than 92% of Fortune 500 companies are building on the platform. ChatGPT now has roughly 100 million weekly active users, primarily growing through word of mouth as people find it useful and share their experiences with others.
Top 12 New GPT-4 Turbo Features
Let’s dive into the top ten features of the new GPT-4 and GPT-4V over the features you already know from GPT-4 (see this article to learn about those):
#1 – GPT-4 Turbo Pricing
Although GPT-4 Turbo is a much better model than GPT-4, OpenAI announced that they will cut the prices of GPT-4 by a factor of 3x for input tokens and 2x for completion tokens.
So the new pricing starting from today is $0.01 per 1,000 input tokens and $0.03 per 1,000 output tokens for GPT-4 Turbo:
In other words, turn off GPT-4 now and switch on GPT-4 Turbo to pay one third of the costs for better performance. No catch! 🤯
#2 – Increasing Context Length
GPT-4 Turbo now supports an impressive context length of up to 128,000 tokens, significantly increasing from the previous 8K token limit. This allows users to work with extensive documents and long contexts more efficiently. You can even load 300 pages of a standard book and it’s 16 times longer than the previous 8k context! Additionally to allowing a longer context, the model is also much more accurate over a longer context!
#3 – Enhanced Control
Developers have more control over model responses and outputs with new features like JSON Mode, improved function calling capabilities, and the introduction of reproducible outputs. These features give developers greater flexibility and precision when utilizing models in their projects.
#4 – Function Calling Improvements
The improved model GPT-4 Turbo allows a better experience in function calling, e.g., you can now call multiple functions simultaneously. In other words, GPT-4 Turbo has enhanced its abilities to effectively carry out multiple function calls concurrently. This translates to a better user experience by adhering to user instructions more precisely.
#5 – Consistent Output Generation: Reproducible Outputs & Seeds
An essential feature, reproducible outputs, is being introduced in GPT-4 Turbo. This enables developers to have control over the model’s behavior, ensuring consistent outputs by including a seed parameter.
I wrote about the OpenAI seeds feature a couple of days ago:
🧑💻 Recommended: DALL·E 3 Trick: Using Seeds to Recreate the Same Image
#6 – Improved World Knowledge
GPT-4 Turbo now incorporates knowledge up to April 2023, making it more up-to-date and useful for users. Additionally, retrieval functionalities enable developers to access external databases and documents for a comprehensive understanding of various subjects.
#7 – Integration of New Modalities
The API now includes DALL-E 3, which generates images programmatically, GPT-4 Turbo with vision capabilities (supporting tasks like image captions, classifications, and analysis), and a new text-to-speech model that produces natural-sounding audio. These advancements create a richer and more interactive experience for end users.
🧑💻 Recommended: GPT-4 with Vision (GPT-4V) Is Out! 32 Fun Examples with Screenshots
#8 – Information Retrieval in the Platform
With the goal of providing improved world knowledge, GPT-4 Turbo launches with a retrieval feature that allows users to acquire knowledge from external documents or databases for their projects.
#9 – DALL-E 3 Unveiled
OpenAI has recently launched DALL-E 3, the latest and most sophisticated image generation model. This cutting-edge technology can be seamlessly integrated with ChatGPT, allowing for more innovative and creative applications. Some businesses have already started using DALL-E 3 to automatically produce images and designs.
For instance, Coca-Cola has initiated a campaign to let customers create Diwali cards with the help of DALL-E 3. Even better, you can now use images as input into the OpenAI API, not only ChatGPT, which was previously impossible!
#10 – Introduction of New Text-to-Speech Model
OpenAI is also introducing a new text-to-speech model, enabling the conversion of text into remarkably natural-sounding audio. With six preset voices to choose from, users can utilize this feature to develop more intuitive and accessible applications. This new model not only enhances language learning experiences but also empowers voice assistants to be more efficient and useful.
#11 – New Speech Recognition Model Whisper V3
OpenAI also introduces the next version of the open-source speech recognition model Whisper 3. I have already written about its predecessor, Whisper, but this one is much better (once again)! Whisper V3 will come soon to the API.
#12 – GPT-4 Finetuning: Custom Models and GPTs
The new program “Custom Models”, the OpenAI researchers will work closely with companies to help them create, well, custom models for their organizations. The idea is to make GPT-4 highly customizable over time.
The vision is already there — as introduced in another recent article on “Introducing GPTs” that allows us to create GPT variations such as a Creative Writing Coach, Tech Advisor, or Sticker Whiz. Without needing to code!
This is what GPTs look like in your browser:
In fact, during the presentation for developers, Sam Altman introduced this slide with different GPTs, i.e., flavors of little AI agents based on GPT-4 Turbo:
I asked GPT-4 to extract all readable labels, i.e., mini GPTs proposed by OpenAI in this short slide:
The image contains numerous icons with labels underneath them. Here’s a list of all the labels that are identifiable in the image:
- Globe Guru
- Star Tracker
- Golf Tracker
- Meditation Guide
- Study Buddy
- Hiking Helper
- In Style
- Pun Generator
- Code Checker
- Pocket Somme (likely “Pocket Sommelier”)
- Tree Houses
- Math Mentor
- Budget Buddy
- Meditation Guide
- Pixel Artist
- all the jokes
This list includes some readable labels, although a few may be incomplete due to partial visibility in the image. For the whole list, check out my detailed article on GPTs:
🧑💻 Recommended: OpenAI Wants Us to Create GPTs, i.e., Miniature AI Agent Helpers
The big picture is clear: we’re entering an age of complete and utter AI penetration into every aspect and job description imaginable.
If you’re reading this article, you already belong to the top 1% of the top 1% of people worldwide who can help make this happen — and benefit the most from the upcoming disruptions.
To learn the ins and outs of prompt engineering and stay on the right side of change, check out our new prompt engineering courses on the Finxter Academy:
While working as a researcher in distributed systems, Dr. Christian Mayer found his love for teaching computer science students.
To help students reach higher levels of Python success, he founded the programming education website Finxter.com that has taught exponential skills to millions of coders worldwide. He’s the author of the best-selling programming books Python One-Liners (NoStarch 2020), The Art of Clean Code (NoStarch 2022), and The Book of Dash (NoStarch 2022). Chris also coauthored the Coffee Break Python series of self-published books. He’s a computer science enthusiast, freelancer, and owner of one of the top 10 largest Python blogs worldwide.
His passions are writing, reading, and coding. But his greatest passion is to serve aspiring coders through Finxter and help them to boost their skills. You can join his free email academy here.