What Paid ChatGPT Offers, as of October 14, 2023

Paid-version features

The paid version includes the following features: GPT-4, a more capable language model; a vision model that can accept images as input; web browsing, which searches for up-to-date information and answers based on it; Advanced Data Analysis, which can analyze uploaded Excel files; plugins, which function like apps that enable ChatGPT to perform various actions; DALL-E 3, the latest image-generation model; and voice conversations, which allow spoken interaction without typing.

GPT-4 versus GPT-3.5

GPT-4 is generally slower than GPT-3.5, but the quality of its answers is much better. The difference is especially clear in Korean.

For academic writing, I recommend the paid version. It can handle up to 4,000 tokens, twice the length of the free version, and there is a major difference in reasoning ability. For requests requiring more sophisticated thinking, such as “Please revise this” or “Please give me review feedback,” the difference is very large.

Vision model (GPT-4V)

The vision model can accept image files as input. In the mobile app, you can take a photo directly and ask questions about it.

I have posted a review of GPT-4V’s capabilities on Twitter: [link].

For academic writing, it is possible to provide a figure and ask it to write a legend. I have received satisfactory results.

Web browsing

A commonly cited limitation of ChatGPT is that it only knows information up to September 2021. When web browsing is enabled, however, it can gather information through Bing search and generate answers based on those results.

For example, I asked it to compare the 2023 revenues of OpenAI, Kakao, and Naver. Previously, it would have said that it did not know. With browsing enabled, it summarizes the search results. Clicking a reference number opens the source page.

For some reason, its search speed is noticeably slower than Bing Chat.

Advanced Data Analysis

I think this is a revolution that will completely change the trend of data analysis. It should make analysis possible even for people who are not familiar with coding or statistics.

For example, I uploaded a CSV file and made only the vague request, “Give me insights.” It analyzed the data comprehensively and even performed regression analysis. This was the result of simply continuing to answer “yes” when it asked, “Shall we continue with additional analysis?”

You can also inspect the Python code used for the analysis. This makes it possible to verify whether the analysis was done properly or copy the code for your own use.

I also asked it to conduct the most creative analysis possible with the data and visualize the results. When I said the colors were difficult to see, it immediately changed the graph colors and redrew it.

Plugins

As of October 14, 2023, there are as many as 998 plugins.

There are many tools that can connect to APIs and carry out tasks, such as booking airline tickets or creating shopping lists. It is possible to connect with Zapier for automation, use Wolfram for calculations or real-time numerical information, or use ShowMe to create charts.

One example request is: “Please draw a chart of OpenAI’s business history using ShowMe.”

Because there are so many plugins, you need to look for and test whether one is suitable for the task you want to perform.

DALL-E 3

When you ask it to draw an image, it uses DALL-E 3, the latest image-generation model.

Even if you do not know advanced prompting techniques, a simple request is automatically converted into an image-generation prompt, so it tends to produce high-quality results.

I expect to use it frequently in the future to create images without copyright concerns for lecture materials and blog posts.

Voice conversations

When voice conversations are enabled in the mobile app, you can speak with ChatGPT without typing, much like in the film Her.

Its speech, including intonation, is surprisingly natural.

It seems very useful for foreign-language speaking practice. If you provide context and instructions at the start—for example, “Correct me when I make grammar mistakes, then move naturally on to the next part of the conversation”—it may be better than many language-learning apps.

I asked it to tell my children a bedtime story about a Pokémon adventure. I asked it to structure the story in 10 chapters, give three choices between chapters, and let the story change based on those choices. It created such an exciting story that it was almost exhilarating. [Twitter link]

Here is the prompt I used: “You are going to tell us a bedtime story about Pokemons going out on an adventure. Make it 10 chapters long, and between each chapter give us 3 option choices. Wait for an answer from the user and adjust the following chapter according to the choice we make. Before starting the first chapter, ask the user who the main character for today’s bedtime story wil be in an open question form.”

Personal note

This is a personal review based on my own purchase and use. I received no support from OpenAI; I paid for it myself and wrote this because I liked it.