What is GPT-4?
GPT-4 is a natural language processing AI model developed by OpenAI that generates natural human-like conversations.
It is a text-based chat AI that provides the right answers to users' questions and requests.
As of 2024, GPT-4 Turbo is the latest version of ChatGPT.
GPT-4 fees
The GPT-4 and GPT-4 Turbo are available for ChatGPT Plus subscriptions of $20 per month (approximately 2,900 yen in Japanese yen).
The API service is credited for the first $18, exceeding which is a pay-as-you-go fee.
API fees for GPT-4, GPT-3.5 and GPT-4 Turbo are as follows.
| GPT-4 Turbo | GPT-4-32k | GPT - 4 | GPT-3.5-turbo-16k | GPT-3.5-turbo-4k | GPT-3.5-turbo | |
| Input fee | $0.01 / 1000 tokens | $0.06 / 1000 tokens | $0.03 / 1000 tokens | 0.001 Dollar/1000 Token | 0.0015 USD / 1000 Tokens | $0.003 / 1,000 talk |
| Output fee | $0.03 / 1000 tokens | 0.12 USD / 1000 Tokens | $0.06 / 1000 tokens | 0.002 USD / 1000 Tokens | 0.002 USD / 1000 Tokens | $0.006 / 1,000 talk |
How is GPT-4 different from GPT-3.5 and GPT-4 Turbo?
GPT-4, GPT-3.5, GPT-4 Turbo, the difference between each is the number of parameters, the amount of training data, the processing capacity, etc.
Below, I will explain in great detail.
Increased number of parameters
The GPT-4, GPT-4 Turbo developed by OpenAI boasts more than five times the number of parameters than GPT-3.5.
With an estimated 1.76 trillion parameters, it has also improved its expressiveness as an enhanced version of GPT-3.5.
GPT-3.5 has 355 billion parameters and was the largest language model available before GPT-4 and Turbo.
The size of these GPT models shows that OpenAI is using a very large data set to train the models.
Answers that fit the user's intent
Compared to the previous version, GPT-4 and GPT-4 Turbo, the tendency to produce inaccurate information, so-called "hulcination", has been greatly reduced.
Being trained on a much broader set of data than GPT-3.5 allows for a deeper understanding of more complex questions and significantly improves the integrity of detailed and relevant answers.
As a result, it is easier to provide answers that match the user's intentions, and more reliable information exchange can be expected.
High-accuracy Japanese language support
Until GPT-3.5, there was a limit to the accuracy of expressions in Japanese, but GPT-4 GPT-4 Turbo has undergone a significant evolution in this regard.
OpenAI's performance tests show a high level of Japanese language proficiency, surpassing the previous model's score in English, resulting in a more fluent and natural language, as well as a more flexible language understanding and expressive ability.
The Japanese spelling, grammar, and accurate understanding of the context have been improved, enabling higher levels of response, from daily dialogue to specialized content.
Ability to solve complex problems
GPT-4 GPT-4 Turbo also surpasses GPT-3.5 in its ability to solve complex problems.
In particular, the accuracy and efficiency of programming-related code generation and writing that requires logical reasoning have been greatly improved.
This means you can rely on AI assistants in the areas of technical documentation, software development, and data analysis.
Not only in the use of private services, but also in the provision of solutions to business and academic issues, the scope of their application is expanding.
Responding to information up to April 2023
GPT-4 and GPT-4 Turbo have different learning data from GPT-3.5.
As GPT-3.5 is learning on data until January 2022, it cannot respond to events or new information since February 2022.
Meanwhile, GPT-4 and GPT-4 Turbo are learning information until April 2023.
This was learned with an update made in late October 2023.
However, no further data has been learned, so you should refer to the latest sources separately for the latest information.
A long sentence of prompt (prompt) is possible
The GPT-4 and GPT-4 Turbo were capable of dealing with long sentences of up to 25,000 characters, allowing for abstracts and translations without splitting long documents.
It has become possible to load a whole book of long documents and books with more than 300 pages, and the range of utilization has expanded greatly.
GPT-3.5 supports up to 5,000 characters.
Multimodal for text and image input
GPT-4 and GPT-4 Turbo have multi-modal functions, which can handle not only text, but also image input.
This made it possible to analyze the images provided by the user and answer the relevant questions.
For example, as shown in the image, the material of the table, the shape and color of the vase, and the figurine tell us in detail that it is a bird.
What you can do with GPT-4
GPT-4 and GPT-4 Turbo can be used in a wide range of fields, including customer support, purchase support, entertainment, AI assistants, data analysis, and marketing.
Leverage versatility and advanced text generation capabilities, including automated customer response, product recommendations, interactive entertainment, daily task management, trend analysis, and content creation to automate and streamline operations.
However, the output speed is inferior to GPT-3.5, and dedicated data analysis tools are suitable for advanced statistical processing and big data processing.
Image generation by DALL E 3
DALL‐E3, an image generation AI service published on September 21, 2023, is available on GPT-4 and GPT-4 Turbo through Bing Image Creator.
Access DALL-E 3 for free through your Microsoft account, with 40 messages per 3 hours.
Customize ChatGPT with GPTs
GPTs is a feature that combines unique data and skills to create your own original GPT for users.
Chat GPT Plus users with GPT-4 and GPT-4 Turbo available can take advantage of this feature.
Documents and files can be provided to the base ChatGPT and customized according to your own instructions.
Customized GPTs can be shared with others or even sold for profit.
No special knowledge is required, and customization work can be done in chat format.
Summary
ChatGPT's GPT-4 and GPT-4 Turbo are interactive AIs developed by OpenAI, which have higher accuracy response and work efficiency capabilities than GPT-3.5.
It is also excellent for complex problem solving, with output accuracy exceeding GPT-3.5, but has a gradual output speed.
Therefore, depending on the type of prompt, it should be used separately from GPT-3.5.
When you understand the features of GPT-4 GPT-4 Turbo, you can use the paid version of ChatGPT more conveniently.
In addition, you can also take advantage of GPT-4 with free services such as Bing AI, other than ChatGPT, so please try it.