GPT-4: OpenAI's Latest Multimodal Natural Language Model
- February 8, 2023
Since the release of GPT-3, especially the ChatGPT based on the GPT-3.5 model, people are highly anticipating the release of GPT-4. With the already impressive performance of GPT-3.5, what level will GPT-4 reach and will its language ability surpass human language capacity? Currently, there is limited official information available about GPT-4. This article will provide the most objective facts.
What is GPT-4
GPT-4 is a new generation of multi-modal large-scale natural language model developed by OpenAI. It can accept image and text input and output accurate text responses. Compared with GPT-3.5, GPT-4 is more reliable and can handle more subtle instructions. On March 14, 2023, OpenAI released an announcement about GPT-4 on its official website. The official said, “GPT-4 is the world’s first advanced AI system with high experience and powerful capabilities. We hope to promote it soon to everyone”.
Price: Start $0.03 / 1K tokens
Tag: AI Chatbot
Release time: March 14, 2023
Developer(s): OpenAI
Share GPT-4
Key Information About GPT-4
- It is planned to be released in 2023 provided that OpenAI deems it to be secure enough.
- It will be stronger than GPT-3 and may potentially reach a new level and bring about a breakthrough (potentially).
- GPT-4 is not AGI, and OpenAI does not currently have AGI.
- On March 9th, 2023, the head of Microsoft Germany announced that GPT-4 will be released next week, and it will be a universal multimodal large-scale model.
- On March 14, 2023, OpenAI launched GPT-4, and also open-sourced OpenAI Evals for automatic evaluation of AI model performance.
- OpenAI released a technical report and System Card, directly upgraded ChatGPT to the GPT-4 version, and also opened the GPT-4 API.
- After the release of GPT-4, Microsoft’s marketing director said that the previous new Bing preview version has used the GPT-4 model.
Rumors Related to GPT-4
Training the model with 100 trillion parameters: Sam Altman has stated this information to be “complete nonsense”.
GPT-4 Function
GPT-4 Enhanced Features
- Give you a more accurate response.
- Text input limit upgraded to 25,000 characters.
- Diversified styles, able to generate lyrics, creative text, etc.
New Features Added to GPT-4
- Superior Academic Ability: GPT-4 performs outstandingly on a variety of professional tests and academic benchmarks, on par with humans.
- Excellent Image Recognition Ability: You can ask questions with text and images, and GPT-4 will give you detailed answers, as if a real human is talking to you.
- GPT-4 API: If you want to access the GPT-4 API, you need to join the waiting list for access.
GPT-4 Currently Pending Features
Image Input: OpenAI is working with other companies to expand the availability of this function.
GPT-4 Limitations
- OpenAI said that GPT-4 still generates wrong answers and makes inference errors, and is not completely reliable.
- GPT-4 reflects a wide range of user values, and various biases may be present in its output.
- GPT-4 generally lacks knowledge of events that occurred after the vast majority of data cutoff (September 2021), and does not learn from experience.
How to test and experience GPT-4
- To access the GPT-4 API (which uses the same ChatCompletions API as gpt-3.5-turbo), you can first register and wait for access, and OpenAI will invite some developers to experience it first.
- Once you have access, you can make plain text requests to the GPT-4 model (Note: image inputs are still in limited alpha).
GPT-4 API Pricing
| Plan | Basic | Pro | ||
|---|---|---|---|---|
|
Price |
$0.03 / 1K prompt tokens |
$0.06 / 1K completion tokens |
$0.06 / 1K prompt tokens |
$0.12 / 1K completion tokens |
|
Function |
Functional for normal use: Output plain text (image input is still in limited alpha) |
Limited access to 32,768 token context (about 50 pages of text) versions |
||
The Development of GPT
- On June 11, 2018, OpenAI released a paper titled “Improving Language Understanding by Generative Pre-Training,” in which they introduced the Generative Pre-trained Transformer (GPT).
- On February 14, 2019, OpenAI first announced GPT-2.
- On May 28, 2020, OpenAI first mentioned GPT-3 in a paper.
- On November 30th, 2022, OpenAI fine-tuned its successor model to GPT-3, called GPT-3.5, to create a brand new application called ChatGPT, which is free for general users to use and rapidly became popular worldwide.
- GPT-4 will be publicly announced at some point in 2023.
- On February 8th, 2023, Microsoft released a new version of Bing based on GPT-4.
- On March 9th, 2023, the head of Microsoft Germany announced that GPT-4 will be released next week.
- On March 14, 2023, OpenAI officially released GPT-4.
