GPT-4: What It Is, How It Works and Differences With Chat-GPT

Finally, GPT 4 or Chat GPT-4 has been released. Since its official launch, the new artificial intelligence technology has generated a lot of buzz about its new features and differences to Chat GPT-3.5. We review the new capabilities of Chat GPT-4 and explain how to use it.

Chat GPT-4, OpenAI's new artificial intelligence chatbot that enhances its older version, Chat GPT-3.5, was officially released on Tuesday 14 March. OpenAI showcased the capabilities of the new AI model in a live stream on YouTube. 

According to the company that created the AI model, Chat GPT-4 is faster, more self-sufficient, more creative and more stable than its previous version: "A year ago, we trained GPT-3.5 as a first 'test' of the system. We found and corrected some bugs and improved our theoretical foundations. As a result, our training run of GPT-4 was of unprecedented stability, becoming our first large model whose training performance we could accurately predict ahead of time."

What is Chat GPT-4?

GPT-4 or Chat GPT-4 is a large multimodal model that leverages the capabilities of artificial intelligence and natural language processing (NLP) to provide natural language answers to any question, in addition to many other capabilities such as writing code, translating and proofreading texts, writing articles, creating spreadsheets, etc. GPT stands for Generative Pre-trained Transformer, a deep learning technology that uses artificial neural networks to write like a human.

According to OpenAI, Chat GPT-4's capabilities closely resemble those of humans: "although less capable than humans in many real-world scenarios, it shows human-level performance in several professional and academic benchmark tests".

How to use Chat GPT-4?

GPT-4 is currently only available in the paid ChatGPT Plus subscription. The current free version of ChatGPT will continue to be based on GPT-3.5. 

GPT-4 will also be available as an API for developers to use for creating applications and services. 

The easiest way to start using GPT-4 for free is to try it as part of Bing Chat. After announcing its partnership with OpenAI, Microsoft has confirmed that its free-to-use browser chatbot, Bing Chat, is already using GPT-4. 

For users who have subscribed to ChatGPT Plus, the way to use GPT-4 Chat is identical to GPT-3.5. Simply tell it what you need and wait a few seconds for the algorithm to respond.

GPT-4: New capabilities

According to OpenAI, GPT-4 is more advanced than its predecessors in three key areas: creativity, visual input and contextual range.

In terms of creativity, OpenAI claims that GPT-4 is much better at both creating and collaborating with users on creative projects. Examples include music, scripting, technical writing and even learning a user's writing style.

By breadth of context, OpenAI refers to the model's ability to process text. GPT-4 can now process up to 25,000 words of text from a user. In addition, users can also send GPT-4 the link to a web page and ask it to interact with the text on that page without having to type it themselves. This new functionality optimises the algorithm's content creation capabilities, as well as its ability to carry on long conversations.

Another major new feature of GPT-4 is image processing. GPT-4 does not only process text, but it is now also capable of interacting with images. That is, we can ask Chat GPT-4 questions or requests by sending an image. 

Finally, OpenAI claims that GPT-4 is much safer than Chat GPT-3.5. According to OpenAI's internal testing, GPT-4 produces 40% more objective responses and is 82% less likely to "respond to requests for inappropriate content". The company behind the AI model claims to have worked with more than 50 experts to get their initial feedback in areas such as artificial intelligence safety and security.

The limitations of Chat GPT-4

After announcing the model and explaining its new capabilities, OpenAI has also addressed some of the limitations of the new AI language model. Like previous versions of GPT, OpenAI claims that the latest model still has problems with "social biases, hallucinations and adverse cues".

By hallucinations, OpenAI refers to the false answers that GPT-3 and GPT-3.5 used to make up when they did not have an answer to the user's question. This is one of the areas that researchers have been working hardest to solve, and they claim that Chat GPT-4 is 40% more likely to give truthful answers than GPT-3.5. 

However, users should be aware that Chat GPT-4 is not perfect and, like its previous versions, may fall into inaccuracies.

Differences between Chat GPT-4 and Chat GPT-3.5

While OpenAI has not explicitly disclosed what changes they have made to the new artificial intelligence model, they have highlighted how much it has improved over Chat GPT-3.5 and described the model's new features. 

In terms of speed and processing power, OpenAI claims that Chat GPT-4 can process up to 25,000 words at a time —eight times more than the original GPT-3 model— and can understand much more nuanced instructions, requests and questions than GPT-3.5, the model used in the existing ChatGPT AI.

OpenAI also claims that GPT-4 is smarter than its predecessor and is less likely to give false answers or fall into inaccuracies. In addition, they claim that it is much harder to fool and more resistant to inappropriate requests.

In addition to the optimisations, here are some of the new features introduced by Chat GPT-4.

The new capabilities of Chat GPT-4

Chat GPT-4 understands images

GPT-4 is a multimodal language model and therefore it does not only understand text, but is also able to process and interact with images and other multimedia elements. 

One of the most celebrated new features of GPT-4 is that it now accepts images as queries and has the ability to answer questions about an image, to generate captions and to classify and analyse images.

It is much better at programming

ChatGPT-3 had programming skills and was already capable of generating code and transforming natural language into code, among other functionalities. However, Chat GPT-4 takes these capabilities to a new level and is a much better programmer than its predecessor. 

One of the most talked-about news concerning Chat GPT-4 in recent days is that the software was able to create and program the game Pong from scratch at the request of a user who had no programming skills.

Chat GPT-4 does pass difficult exams

One of the main complaints about Chat GPT was linked to its inability to pass relatively easy exams. When the Chat GPT revolution exploded, several teachers and professors decided to test the software's intelligence to see if it really was as smart as a human. Many users reported that, according to their testimony, Chat GPT-3 was incapable of passing a primary or secondary school exam.

However, according to OpenAI and users who have already tested the algorithm, Chat GPT-4 performs much better if you put it to the test. GPT-4 has already passed Uniform Bar —one of the most complex bar exams— ranking in the 90th percentile of test takers. It has also passed the Biology Olympiad test in the 99th percentile. By comparison, ChatGPT could only do so in the 31st percentile. 

It has a better sense of humour

Compared to the old version of GPT Chat, GPT-4 understands much better what makes something funny. It tells better jokes and if you show it a meme or other funny picture and ask it to explain what's funny about it, it can understand what's going on and explain it to you.

Conclusion

Chat GPT-4 promises to be an improved version of the deep learning chatbot that has caused the most excitement in recent years. With new features and improved capabilities, the model takes us a step further in the field of artificial intelligence. However, the company behind the model itself admits that Chat GPT-4 is not perfect and still has its flaws.