What's different about GPT-4?

Woke up and GPT-4 was released.

GPT-4 and GPT-3.5 are both large-scale natural language processing models developed by OpenAI that are capable of understanding and generating human language. Although the two models are similar in many ways, there are some key differences between them:

1. More creative

Although the expression of GPT-3.5 is already amazing , but GPT-4 is more creative and collaborative.

GPT-4 has more parameters than GPT-3.5, which makes GPT-4 more capable at handling complex tasks and generating more accurate and natural answers. This increased capability results in greater accuracy and better quality text generation.

The example from the official website is very interesting:

Use one sentence to explain the plot of Cinderella. Each word must start with the next letter of the alphabet from A to Z without repeating any letter. The answer given in GPT-4 is:

A beautiful Cinderella, dwelling eagerly, finally gains happiness; inspiring jealous kin, love magically nurtures opulent prince; quietly rescues, slipper triumphs, uniting very wondrously, xenial youth zealously.

The result in 3.5 is like this, the gap is still quite big:

2. Multi-modality

To put it simply, in GPT-3.5 you can only input text, but in GPT-4 you can input text and pictures at the same time and output text. It can understand the information in the picture and give answers based on the input text. For example, give a landscape photo and let GPT-4 tell you where it is and make a travel plan.

I believe that audio and video will be supported soon, which will create more possibilities.

3. Support for long text

GPT-4 is capable of handling text in excess of 25,000 words, allowing for applications such as long-form content creation, extended sessions, document search, and analysis.

You can even directly ask him to read the content of a link and summarize the content:

4. Timeliness

The last training time of GPT-4 is September 2021. It feels similar to GPT-3.5. I thought it was the latest data at first.

Despite this, GPT-4 knows itself better, and those who don’t know will not talk nonsense like GPT-3.5. For example, in GPT-3, I asked me to rate the movie Man Jiang Hong:

In GPT-4, the results are more objective:

##5. Security

Before GPT- 3.5 often outputs bad content under some inducements. In this regard, GPT-4 performs better. It is 82% less likely to respond to disallowed content than GPT-3.5 and 40% more likely to produce factual content. respond.

To experience GPT-4 now, there are the following ways:

1. Upgrade to ChatGPT Plus on the official website of openai, which is a bit expensive:

2. poe.com has integrated GPT-4, but there are usage restrictions. You have to wait a few hours after using it once:

The above is the detailed content of What's different about GPT-4?. For more information, please follow other related articles on the PHP Chinese website!