GPT-4: all about the new chatbot

OpenAI has just unveiled the most recent version of its language models, GPT-4. The previous version, GPT-3.5, is the one that has powered ChatGPT since its launch in November 2022.

GPT stands for Generative Pre-trained Transformer. It is a type of language model that uses deep learning to generate human-like conversational text.

What can GPT-4 do?

Since GPT-4 is a broad multimodal model, it is able to accept both text and images as input and produce human-like text.

For example, if you scan a spreadsheet with GPT-4, it will be able to scan it and output the answers to the associated questions. The model can also read a graph and perform calculations based on the data presented.

The intellectual capabilities of the model have also been improved in this version, which outperforms GPT-3.5 in a series of mock benchmark exams, as shown in the graph below (with examples of university degrees).

ChatGPT-4 capabilities

Screenshot by Sabrina Ortiz/ZDNET.

When was GPT-4 launched?

GPT-4 was officially unveiled by OpenAI on March 14, 2023, almost four months after ChatGPT launched to the general public.

How to access GPT-4?

OpenAI has yet to make GPT-4’s visual input capabilities available on any platform, as the company is working with a single partner to begin with. However, it is already possible to use GPT-4 via text input.

And that access through OpenAI comes with a subscription to ChatGPT Plus, which guarantees users access to the new language model for $20 a month. However, even with this subscription, there is a request limit, which means that you may not be able to access it all the time, something to consider before making any investment.

However, there is a way to access GPT-4 text input for free, using Bing Chat. On the day GPT-4 was unveiled by OpenAI, Microsoft said its own chatbot, Bing Chat, had already been running on GPT-4 since its launch five weeks ago. The use of Bing Chat is free, after registration on a waiting list.

What is Bing Chat?

Bing Chat is Microsoft’s chatbot. It runs on OpenAI’s most advanced LLM, GPT-4. A month after its launch, 45 million chats have taken place on the platform and the new Bing has 100 million daily active users (DAUs), according to Yusuf Mehdi, Microsoft vice president and chief consumer marketing officer.

The popularity of the chatbot is explained by the fact that it has the same capabilities as ChatGPT, but it has full access to the internet, which is not the case with ChatGPT (used on the site of Open AI). And for now, the public can only use it by registering on a waiting list.

Is there a GPT-4 API?

Yes, a GPT-4 API is available for developers through a waiting list.

The waiting list asks for specific information about how you plan to use the GPT-4. Like, for example, building a new product, integrating into an existing product, academic research, or just general exploration of GPT-4’s capabilities. The form also asks you to share any specific ideas you have for using GPT-4.

What model does ChatGPT currently use?

ChatGPT is powered by GPT-3.5, which limits the chatbot to inbound and outbound text exchanges.

What is the difference between GPT-4 and GPT-3.5?

The main difference between the models is that GPT-4 being multimodal, it can use images in addition to text, whereas GPT-3.5 can only handle text.

According to OpenAI, the distinction between GPT-3.5 and GPT-4 will also be more “subtle” in ordinary conversation. However, the new model would perform better in terms of reliability, creativity and even intelligence, as evidenced by the higher performance obtained in the benchmark exams mentioned above.

Can GPT-4 still be wrong in its answers?

Yes, GPT-4 is subject to the same limitations as previous GPT models. OpenAI even indicates that this model is “not completely reliable (it “hallucinates” facts and makes errors of reasoning)”.

Despite this warning, OpenAI claims that GPT-4 hallucinates less often than previous models. The latest version of the LLM has indeed achieved significantly higher scores than GPT-3.5 in an internal assessment made in several knowledge areas, as shown in the table below.

ChatGPT internal factual evaluation by category

Screenshot by Sabrina Ortiz/ZDNET.

Source : ZDNet.com

Disclaimer: If you need to update/edit/remove this news or article then please contact our support team Learn more

Varun Kumar

Varun Kumar is a freelance writer working on news website. He contributes to Our Blog and more. Wise also works in higher ed sustainability and previously in stream restoration. He loves running, trees and hanging out with her family.

Leave a Reply