Related Links
OpenAI official website, click me https://openai.com/
gpt1234, chatGPT are made by them, the official website has a lot of information, you can go here to find it directly.
GPT-4 official introduction, click me https://openai.com/product/gpt-4
GPT-4 research introduction, click me https://openai.com/research/gpt-4
how to use
Open chatGPT plus, about 100 yuan a month
or join the waiting list
gpt-4 product introduction
GPT-4 is OpenAI's state-of-the-art system that produces safer, more useful responses. GPT-4 can solve hard problems more accurately, thanks to its broader common sense and problem-solving abilities.
This is the latest milestone in OpenAI's efforts to scale deep learning. GPT-4 is a large multimodal model (accepts image and text input, emits text output) that, while inferior to humans in many real-world scenarios, demonstrates human-level performance on various professional and academic benchmarks.
creativity
GPT-4 is more creative and collaborative than ever. It can generate, edit, and work with users on creative and technical writing tasks, such as composing songs, writing scripts, or learning the user's writing style.
Input
Explain the plot of Cinderella in a sentence where each word has to begin with the next letter in the alphabet from A to Z, without repeating any letters. Output
A
beautiful Cinderella, dwelling eagerly, finally gains happiness; nurses opulent prince; quietly rescues, slipper triumphs, uniting very wondrously, xenial youth zealously.
Enter
explaining the plot of Cinderella in sentences where each word must begin with the next letter of the alphabet from A to Z, without repeating any letter.
Output
Beautiful Cinderella, eager to live in peace, and finally happy; arouses jealous relatives, love magically breeds a rich prince; quietly rescues, slippers win, united, and enthusiastic youth.
visual input
GPT-4 can accept images as input and generate descriptions, classifications, and analysis.
Input
What can I do with these ingredients?
Output
You can do a lot of options with these ingredients! Some possibilities include: - pancakes or waffles - crepes - french toast - omelet or frittata - quiche - custard or pudding - cakes or cupcakes - muffins or bread - Cookies or Cookies These are just a few examples, but the possibilities are endless!
Input:
What's so funny about this picture? Describe it panel by panel.
Output:
This image shows the packaging of the "Lightning Cable" adapter with three panels.
Panel 1: A smartphone with a VGA connector (a large blue 15-pin connector typically used on computer monitors) plugged into its charging port.
Panel 2: Packaging for the "Lightning Cable" adapter with a picture of the VGA connector.
Panel 3: A close-up of the VGA connector with a small Lightning connector on the end (used to charge iPhones and other Apple devices).
The humor in this image comes from the ridiculous practice of plugging an outdated large VGA connector into a small modern smartphone charging port.
longer context
GPT-4 is capable of processing over 25,000 words of text, allowing use cases such as long-form content creation, extended conversations, and document search and analysis.
Input
[Text from: https://en.wikipedia.org/wiki/Rihanna]
Describe her Super Bowl performance and what made it special.
Output
Rihanna's Super Bowl LVII halftime show performance was special for a number of reasons. For one, it marked her first live performance in over five years, and it was her first time headlining the Super Bowl halftime show. Additionally, she made history by becoming the first person to perform at the Super Bowl while pregnant. The show itself was critically accclaimed, with Many praising her high-energy performance and impressive vocals. It also garnered a massive viewership, becoming the second most-watched halftime show in history.
GPT-4 surpasses ChatGPT in advanced reasoning
ChatGPT
Input
Andrew is free from 11 am to 3 pm, Joanne is free from noon to 2 pm and then 3:30 pm to 5 pm. Hannah is available at noon for half an hour, and then 4 pm to 6 pm. What are some options for start times for a 30 minute meeting for Andrew, Hannah, and Joanne?
Output
Here is a list of their availability:
- Andrew: 11 am to 3 pm
- Joanne: noon to 2 pm, and 3:30 pm to 5 pm
- Hannah: noon to 12:30 pm, and 4 pm to 6 pm
Based on their availability, there is a 30-minute window where all three of them are available, which is from 4 pm to 4:30 pm. So, the meeting can be scheduled at 4 pm.
GPT-4
Input
Andrew is free from 11 am to 3 pm, Joanne is free from noon to 2 pm and then 3:30 pm to 5 pm. Hannah is available at noon for half an hour, and then 4 pm to 6 pm. What are some options for start times for a 30 minute meeting for Andrew, Hannah, and Joanne?
Output
Andrew: 11 am - 3 pm
Joanne: 12 pm - 2 pm, 3:30 pm - 5 pm
Hannah: 12 pm - 12:30 pm, 4 pm - 6 pm
Common availability for a 30-minute meeting: 12 pm - 12:30 pm
GPT-4 scored higher than ChatGPT in professional exams
Uniform Bar Exam:
ChatGPT 10 points
GPT-4 90 points
Biology Olympiad:
ChatGPT 31 points
GPT-4 99 points
| Exam, Model, Score | ChatGPT | GPT-4 |
|---|---|---|
| Uniform Bar Examination | 10 | 90 |
| biology olympiad | 31 | 99 |
GPT-4 landing products
GPT-4 deepens conversations on Duolingo
Be My Eyes uses GPT-4 to transform visual accessibility
Stripe leverages GPT-4 to simplify the user experience and fight fraud.
Morgan Stanley Wealth Management deploys GPT-4 to organize its vast knowledge base
Iceland uses GPT-4 to protect its language
Khan Academy explores potential of GPT-4 in limited pilot program
training process
Like previous GPT models, the GPT-4 base model is trained to predict the next word in a document and is trained using publicly available data such as internet data as well as data we have licensed. These data are web-scale corpora of data that include correct and incorrect solutions to mathematical problems, weak and strong reasoning, contradictory and consistent statements, and represent a wide variety of ideologies and ideas.
Thus, when prompted with a question, the underlying model can respond in a variety of ways that may be far from the user's intent. To align it with user intent within the guardrails, we use reinforcement learning and human feedback (RLHF) to fine-tune the behavior of the model.
Note that the model's power seems to come mostly from the pre-training process - RLHF does not improve test scores (it actually decreases test scores if not actively worked on). But the control of the model comes from the post-training process - the underlying model needs to be engineered in time to even know it's supposed to answer the question.
limit
Although powerful, GPT-4 has similar limitations to earlier GPT models. On top of that, it's still not entirely reliable (it "hallucinates" facts and makes inference errors). Care should be taken when using language model output, especially in high-stakes contexts, using the exact protocol that matches the needs of the specific use case (e.g. human review, a basis for additional context, or avoiding high-stakes use altogether). While still
a Real problem, but GPT-4 significantly reduces hallucinations relative to previous models (which themselves improved with each iteration). In our internal adversarial realism evaluation, GPT-4 scores 40% higher than our latest GPT-3.5:
More information
Research: GPT-4 is the latest milestone in OpenAI's effort to scale deep learning. Check out the GPT-4 research
Infrastructure: GPT-4 was trained on a Microsoft Azure AI supercomputer. Azure's AI-optimized infrastructure also enables us to deliver GPT-4 to users around the world.
Limitations: GPT-4 still has many known limitations that we are working to address, such as social bias, hallucinations, and adversarial cues. As society adopts these models, we encourage and promote transparency, user education, and broader AI literacy. We also aim to expand the avenues of input for people in shaping our models.
Availability: GPT-4 is available on ChatGPT Plus and serves as an API for developers to build applications and services.
We're excited to see how people use GPT-4 as we work to develop technology that empowers everyone.
in conclusion
We look forward to GPT-4 becoming a valuable tool that improves people's lives by powering many applications. There is still much work to be done, and we look forward to improving the model through the collective efforts of the community to build, explore, and contribute on top of it.