OpenAI has launched GPT-4, a powerful new AI model designed for image and text comprehension, representing the company’s latest step in scaling up deep learning. The model is accessible to OpenAI’s paying customers through ChatGPT Plus, with usage caps, and developers can apply to join a waitlist to use the API.
GPT-4’s pricing is based on tokens, with $0.03 per 1,000 prompt tokens, which are the portions of words fed into the model, and $0.06 per 1,000 completion tokens, which are the content generated by GPT-4. Each prompt token contains about 750 words and is used as a raw text input for the model to produce relevant content through completion tokens. This launch marks a significant advancement in the field of AI technology and demonstrates OpenAI’s commitment to expanding its deep learning capabilities.
The Launch of ChatGPT-4 With Bing Chat
Microsoft has recently revealed that its Bing Chat, a chatbot technology co-developed with OpenAI, is operating on GPT-4. Interestingly, GPT-4 has already been adopted by several early adopters, including Stripe, which utilizes it to scan business websites and provide a summary to its customer support team. Duolingo has integrated GPT-4 into a new language learning subscription tier, while Morgan Stanley is developing a system powered by GPT-4 to extract information from company documents and present it to financial analysts.
Additionally, Khan Academy is utilizing GPT-4 to create an automated tutor. These early adopters demonstrate the potential of GPT-4 to revolutionize various industries and highlight the rapid advancements in AI technology. The use of GPT-4 in a range of applications shows its versatility and marks a significant step forward in the development of intelligent machines capable of performing human-like tasks.
ChatGPT 4 Predecessor- ChatGPT 3.5
Compared to its predecessor, GPT-3.5, OpenAI’s latest AI model, GPT-4, demonstrates an improvement by being capable of generating text and accepting both image and text inputs. Furthermore, GPT-4 outperforms GPT-3.5 on various academic and professional benchmarks, exhibiting a “human-level” of performance.
In a simulated bar exam, GPT-4 scores within the top 10% of test-takers, a significant improvement over GPT-3.5, which scored within the bottom 10%. This advancement showcases the impressive capabilities of GPT-4 and the potential impact it can have in various industries that rely on natural language processing and image recognition. The success of GPT-4 in these tests suggests that it could be employed in a range of applications, including legal research, content generation, and customer service.
According to OpenAI, GPT-4 has achieved the “best-ever results” on factuality, steerability, and adherence to guidelines after six months of iterative alignment using insights from an internal adversarial testing program. Data that was both openly accessible (such as that found on public web pages) and licensed by OpenAI were used to train GPT-4. To train GPT-4, OpenAI collaborated with Microsoft to develop a supercomputer in the Azure cloud from scratch.
Although the difference between GPT-3.5 and GPT-4 may be subtle in casual conversation, OpenAI claims that GPT-4 is more reliable, creative, and capable of handling more nuanced instructions than its predecessor when the task becomes complex enough. GPT-4’s ability to understand images, in addition to text, is one of its most intriguing features. For instance, it can add subtitles and even decipher rather sophisticated photographs, such recognizing a Lightning Cable converter from a picture of an iPhone plugged in.
At present, GPT-4’s ability to comprehend images is limited to a single partner, Be My Eyes, and isn’t accessible to all OpenAI customers. Be My Eyes has developed a new feature called Virtual Volunteer, which is powered by GPT-4 and can respond to inquiries about pictures sent to it. The company has provided a detailed explanation of how it operates, noting that it can accurately recognize the contents of a user’s refrigerator and proposes recipes based on the available ingredients.
One of GPT-4’s most significant improvements may be its steerability tooling. OpenAI is introducing a new API capability with GPT-4 called “system” messages that allows developers to describe specific directions and prescribe style and task, setting the tone and establishing boundaries for the AI’s next interactions. System messages will also be available to ChatGPT in the future.
What are Some of the Limitations of ChatGPT4?
OpenAI has admitted that despite the upgrades, GPT-4 still has limitations. It can still make mistakes while generating content, and sometimes, these mistakes are made with high confidence levels. OpenAI has recognized that GPT-4 still “hallucinates” facts and makes reasoning errors and cited an example where GPT-4 described Elvis Presley as the “son of an actor”, which is factually incorrect. OpenAI also admitted that there are other limitations to GPT-4, despite its significant advancements.
However, the company is continuously working to improve the model’s accuracy and minimize errors. They have introduced a new API capability called “system” messages, which allows developers to give specific directions to GPT-4 by prescribing style and task, thus making it easier to steer the AI’s next interactions. Although GPT-4 still has room for improvement, OpenAI hopes that these upgrades will take them one step closer to achieving their goal of developing a machine that can think and reason like a human.
How do I use ChatGPT+ to gain entry to GPT-4?
To easily generate text using the GPT-4 model, you will need to have a paid membership to ChatGPT+. This premium edition of ChatGPT exclusively uses GPT-4, as stated on OpenAI’s GPT-4 announcement website. If you’re already signed in to your OpenAI account, clicking the ChatGPT icon will take you directly to ChatGPT.
However, if you only have a free account, you will need to upgrade to ChatGPT+ by clicking the “Upgrade to Plus” icon on the left-hand side. This will take users to a regular e-commerce checkout page where they will need to input purchase information and pay $20 per month. Once users become paying customers, they will have instant access to the new GPT-4-enabled ChatGPT.
If a user initiates a new chat in ChatGPT, they will be presented with a pulldown menu that gives them the choice of using one of the older models or GPT-4. For those who are already subscribed to ChatGPT, they may be asked to transition to GPT-4. If they do, a text bar will appear in the center of their current chat stating, “This conversation’s prior model is no longer accessible. We’ve updated the preset model for you.”
GPT-4-powered outputs will have a new, black emblem, while older GPT-3-powered outputs will still have the previous, green OpenAI insignia to the left. Users can quickly distinguish between the two models in this way.
While some may be hesitant to pay for ChatGPT+, those who are serious about generating high-quality, coherent text may find it well worth the cost. GPT-4’s language generation capabilities are significantly more advanced than its predecessor, GPT-3, and can be a valuable tool for those who need to generate large amounts of text quickly. With its many new features and improved accuracy, GPT-4 is poised to become a game-changer in the world of language generation.
How to Get Free Access to GPT-4?
The current version of the free ChatGPT doesn’t support GPT-4, but some Bing users can use the GPT-4 enabled Bing chatbot, albeit with some limitations. Microsoft confirmed that the GPT-4 technology is already being used in conversations for a select group of users. Additionally, OpenAI’s recent blog post about the applications of GPT-4 reveals that various clients, including Khan Academy, Stripe, Morgan Stanley, and the Icelandic government, are already using it. This means that whether you are ready or not, you will soon be encountering GPT-4 in your daily life.
According to OpenAI, GPT-4 has some limitations despite its enhancements. The AI model lacks knowledge of events that occurred after September 2021, and it may sometimes make reasoning errors and accept false statements. It can also introduce security vulnerabilities into the code it produces.
However, OpenAI has made improvements in some areas such as GPT-4 being less likely to respond to requests for disallowed content compared to GPT-3.5. Furthermore, the model is better at responding to sensitive requests in accordance with OpenAI’s policies. Despite the limitations, OpenAI is confident in the enhancements it has made and is looking forward to GPT-4 becoming a valuable tool in improving people’s lives. The company believes that with the community’s collective efforts building on top of, exploring, and contributing to the model, they can continue to improve it.