The official release of ChatGPT-4 confirms long-running reports about its enhancements to the OpenAI’s ChatGPT’s already astounding linguistic abilities. GPT-4’s broader general knowledge and problem-solving skills enable it to more accurately resolve complex issues.
GPT-4 is more innovative and team-oriented then ever before. It can generate, edit, and iterate with users on artistic and technical writing tasks, such as producing music, writing screenplays, or learning a user’s writing style. GPT-4 can produce captions, classifications, and analysis using photos as inputs.
GPT-4 can process more over 25,000 words of text, which enables use cases including the generation of long form content, lengthy dialogues, and document search and analysis. It is the company’s “most advanced algorithm, providing safer and more helpful solutions,” according to OpenAI. Here is everything we currently know about it.
Availability
Microsoft indicated in advance that GPT-4 would be publicly introduced on March 13, even though the precise date was unknown. Nevertheless, as of right moment, it’s only accessible through the ChatGPT Plus premium membership. The free version of ChatGPT that is currently available will continue to use GPT-3.5, which is less accurate and competent in comparison.
In addition, an API for GPT-4 will be made available “enabling developers to construct apps and services.” Among of the firms that have already integrated GPT-4 include Duolingo, Be My Eyes, Stripe, and Khan Academy. The first public GPT-4 presentation, showcasing some of its new features, was also livestreamed on YouTube.
How to use GPT-4
Today, using GPT-4 as a part of Bing Talk is the simplest way to get started. Microsoft acknowledged utilising the free-to-use GPT-4 in Bing Talk. However, some GPT-4 capabilities, including visual input, are absent from Bing Chat. Yet you’ll still have access to the additional intelligence that comes with the extended LLM (large language model). It should be noted that although Bing Chat is free, there is a 15-chat session limit and a daily session limit of 150.
At present, upgrading to ChatGPT Plus is the only other option for getting access to GPT-4. Click “Upgrade to Plus” in the ChatGPT sidebar to upgrade to the $20 premium subscription. You can switch between GPT-4 and earlier iterations of the LLM after providing your credit card details. As GPT-4 responses use a black logo rather than the green emblem used for older versions, you can even confirm that you are receiving GPT-4 responses.
The process of utilising GPT-4 after then is the same as using ChatGPT Plus with GPT-3.5. It is more powerful than ChatGPT and lets you adjust a dataset to produce results that are specific to your needs.
What’s new in GPT-4?
The new language model GPT-4, developed by OpenAI, can produce text that sounds like human speech. It advances the GPT-3.5-based technology that ChatGPT currently employs. Generative Pre-trained Transformer, sometimes known as GPT, is a deep learning tool that employs artificial neural networks to write naturally.
This new generation of language models, according to OpenAI, is better advanced in three crucial areas: inventiveness, visual input, and lengthier context. According to OpenAI, GPT-4 is significantly more creative and is much better at working with users on creative projects. They include, for instance, technical writing, music, scripts, and even “understanding a user’s writing style.”
The longer context also affects this. Up to 25,000 words of text from the user can now be processed by GPT-4. You can even ask GPT-4 to interact with text from a web page by simply sending it a link. According to OpenAI, this can be useful for “longer dialogues” as well as the generation of long-form content.
Moreover, GPT-4 may now accept photos as a foundation for communication. The chatbot is shown an image of a few baking ingredients in the example supplied on the GPT-4 website, and it is then asked what can be produced with them. If video may be used in a similar manner is currently unknown.
Last but not least, OpenAI claims that GPT-4 is much safer to use than GPT-3. According to internal testing by OpenAI, it can generate 40% more factual responses while also being 82% less likely to “respond to requests for banned content.”
OpenAI claims to have collaborated with “over 50 experts for early feedback in domains including AI safety and security,” and that this progress has been made through training with human feedback.
We’re starting to discover its potential as the first people have flocked to use it. Using a combination of HTML and JavaScript, one user is said to have managed to get GPT-4 to produce a playable version of Pong in under 60 seconds.
8 Ways …GPT-4 is Impressive – A Comparison with GPT 3.5
An A.I. chatbot that fascinated the IT industry four months ago now has a newer, more advanced version of the technology that runs it. It is an authority on many topics and even dazzles doctors with its medical advice. It’s close to cracking jokes that are almost funny and can explain imagery.
Yet, the long-rumored new AI system, GPT-4, still exhibits some of the peculiarities and commits some of the same repetitive errors that perplexed researchers when that chatbot, ChatGPT, was first released. The system, developed by San Francisco start-up OpenAI, is quite adept at taking tests, but it is not yet at the level of human intelligence. This is a quick overview of GPT-4:
1. It can describe images with impressive detail.
GPT-4 now has the capacity to react to both text and images. The president and co-founder of OpenAI, Greg Brockman, demonstrated how the system could painstakingly explain an image from the Hubble Space Telescope. Several paragraphs of description were given.
It can also respond to inquiries regarding an image. It can propose a few meals to prepare with the ingredients on hand if given a picture of a fridge’s interior.
OpenAI has not yet exposed this component of the technology to the public, but a business called Be My Eyes is already employing GPT-4 to build services that might give a more detailed notion of the images encountered on the internet or taken in the real world.
2. It can give editors a run for their money.
The new chatbot can almost always provide a clear and accurate synopsis of a story when given an article from The New York Times. The bot will point to the added sentence if you add a random sentence to the summary and ask it if the summary is accurate.
That was noted as a great skill by Dr. Etzioni. “It has to have a level of grasp of a text and a capacity to convey that understanding,” he added, “to produce a high-quality summary and a high-quality comparison. “That’s a higher level of intelligence,”
3. It can reason — up to a point.
Dr. Etzioni gave the new bot a puzzle.
The system appeared to react correctly. Nevertheless, the height of the entryway was not taken into account in the solution, which might also make it impossible for a tank or a car to pass through.
Sam Altman, the CEO of OpenAI, said that the new bot had “some” reasoning abilities. But there are many circumstances where its ability to reason fails. Because it understood that height and width mattered, the prior version of ChatGPT handled the question a little better.
4. It has improved its accuracy.
A.I. researcher and professor Oren Etzioni posed a direct query when he first used the new bot: “What is the relationship between Oren Etzioni and Eli Etzioni?” The bot gave the right answer.
The response provided by ChatGPT’s earlier iteration to that query was invariably inaccurate. If the new chatbot answers correctly, it demonstrates a wider breadth of expertise. But it continues to produce errors.
The bot continued, “Eli Etzioni is an entrepreneur, whereas Oren Etzioni is a computer scientist and the CEO of the Allen Institute for Artificial Intelligence (AI2).” The majority of that is true, but the bot, whose training was finished in August, was unaware that Dr. Etzioni had lately left her position as president of the Allen Institute.
5. It has learned to be more precise.
On a recent day, Chris Nicholson, a specialist in artificial intelligence and a partner with the venture capital firm Page One Ventures, utilised GPT-4. He told the bot that he spoke only English and had no knowledge of Spanish.
The bot gave him a thorough and well-organized syllabus when he requested one that could teach him the fundamentals. Even a variety of methods for learning and retaining Spanish words were supplied (though not all of its suggestions hit the mark).
Similar assistance was requested by Mr. Nicholson from ChatGPT’s earlier iteration, which used GPT-3.5. Although it too offered a syllabus, its recommendations were less specific and less effective.
6. It is developing a sense of humor. Sort of.
“A fresh joke about the singer Madonna,” Dr. Etzioni requested of the new bot. The answer pleased him. It also made him giggle. It might also impress you if you are familiar with Madonna’s best singles.
The new bot was still having trouble writing anything besides predictable “dad jokes.” But it was just a little bit funnier than the previous one.
7. It has added serious expertise.
Anil Gehi, a cardiologist and associate professor of medicine at the University of North Carolina at Chapel Hill, spoke with a chatbot recently and provided it with information about a patient he had seen the day before, including complications the patient encountered after being admitted to the hospital. There were various medical terminology in the description that non-medical persons would not understand.
The chatbot provided Dr. Gehi with the ideal response when he questioned how he ought to have handled the case. He answered, “That is exactly how we handled the patient.
He tried different circumstances, and the bot responded with equally astounding responses.
It’s doubtful that this information will always be shown when the bot is in use. It still requires professionals like Dr. Gehi to assess its reactions and perform the necessary medical procedures. Yet, it can demonstrate this level of skill in a wide range of fields, including accounting and computer programming.
8. It can ace standardized tests.
For the Unified Bar Test, which certifies attorneys in 41 states and territories, OpenAI said the new approach might place in the top 10% or so of students. According to the company’s tests, it can also get a 1,300 on the SAT (out of 1,600) and a 5, out of 5, on Advanced Placement high school examinations in biology, calculus, macroeconomics, psychology, statistics, and history.
Earlier iterations of the system failed the Uniform Bar Test and performed far worse on most AP exams. On a recent afternoon, Mr. Brockman sent the new bot a bar exam question about a man who owns a diesel-truck repair business in order to illustrate its testing capabilities.
The solution was right, but it was written in legalese. As a result, Mr. Brockman asked the robot to explain the solution in layman’s terms. That was also done.
Pricing and Enlarged Context Window
An key benefit of the API is that it gives users access to a larger context window. GPT-4 supports prompts up to 8K and 32K tokens (25K words), which translates to documents of up to 50 pages. With GPT-4, some applications that were not practical with GPT-3.5 (such as processing a full book in one or a few passes) are simple. (ChatGPT+ users don’t appear to have access to this option.)
Also, you must spend more if you want to analyse more data simultaneously. This is the pricing scheme for the GPT-4 API:
- 8K tokens: $0.03/1k prompt tokens, $0.06/1k completion tokens.
- 32K context: $0.06/1k prompt tokens, $0.12/1k completion tokens.
For comparison, the next best model, which underlies ChatGPT (i.e., GPT-3.5-turbo), costs $0.002 /1K tokens (15x less than the cheapest option for GPT-4) and doesn’t differentiate between prompt and completion. Depending on your use case, it could make no sense to switch to GPT-4.
Summary
It is obvious that OpenAI’s primary goal for their GPT-4 platform is to develop a partner ecosystem. It is evident from the wide range of initiatives and programmes that people are interested in applying cutting-edge NLP models to many different sectors of society and business. In summary, GPT-4 has a wide range of potential applications, from news writing and content generation to chatbots and automated customer service systems.
However, there are still a lot of recognised issues with GPT-4 that they are trying to fix. “Social biases, hallucinations, and antagonistic cues” are a few of these. OpenAI understands that increasing functionality also entails accountability, thus their team is working to make sure that GPT-4, like its predecessors, maintains a focus on responsible AI.