OpenAI releases GPT-4, which has stronger question-solving ability. It also accepts image input, making it easy to understand memes!

GPT-4 is finally released!

After the capital market carnival, GPT-4 was born under the spotlight. OpenAI released it on its official website with the title "GPT-4 is OpenAI's most advanced system, capable of producing safer and more useful responses" This latest model.

This iteration is not slow at all, as ChatGPT launched just four months ago and has become the fastest-growing consumer application in history.

In general, GPT-4 improves the core technology of ChatGPT, making it have a wider range of general knowledge and problem-solving capabilities than previous versions, allowing the chat software to operate with higher accuracy Solve more difficult problems.

In addition, it also adds some new features, such as accepting images as input, generating titles, classification and analysis.

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

New Ability: Understanding Meme Accessibility

One of the biggest updates in the GPT-4 model is that it introduces images as input and It generates titles, categories, and analyses, whereas GPT-3.5 can only accept text.

For example, the following example:

Question: What is funny about this picture? Describe it bit by bit.

GPT-4 Answer: This image shows a "Lightning Cable" adapter with three pictures.

Figure 1: A smartphone with a VGA connector (a large, blue, 15-pin connector commonly used on computer monitors) plugged into its charging port.

Figure 2: The package of the “Lightning Cable” adapter, with a picture of the VGA connector on it.

Figure 3: Close-up of the VGA connector, ending with a small Lightning connector (used for charging iPhones and other Apple devices).

The humor in this image comes from the absurdity of plugging a large, outdated VGA connector into a small, modern smartphone charging port.

Currently, while GPT-4 is not powerful enough to enable text-to-video generation and other dynamically generated content, it does provide a glimpse of what multi-element chat will look like in the future. It’s easy to imagine how video, audio, images and all other forms of content will be integrated into chats in the future.

Old capabilities: When the task is complex enough, GPT-4 will be a step higher than GPT-3.5

GPT-4, as OpenAI’s latest milestone in expanding deep learning, is essentially A large multi-modal model (accepting image and text input, emitting text output), although inferior to humans in many real-world scenarios, its capabilities have approached or even exceeded human-level performance on various professional and academic benchmarks .

For example, it passed the Mock Bar Exam with scores around the top 10%; in contrast, GPT-3.5 scored around the bottom 10%.

The difference between GPT-3.5 and GPT-4 may be subtle if you are just having a casual conversation, but when the complexity of the task reaches a sufficient threshold, the difference will become apparent - and GPT- Compared with 3.5, GPT-4 is more reliable, more creative, and can handle more nuanced instructions.

To understand the differences between the two models, OpenA tested various benchmarks, including simulating exams originally designed for humans, and it can be seen that in more advanced tests, GPT-4 performed better than GPT -3.5 is much better.

As another example, it can be seen that GPT-4 does have better understanding capabilities.

OMT: Introducing more human feedback

In order to increase the complexity of the model, GPT-4 also adds more human feedback, including ChatGPT users Submit feedback to improve the behavior of GPT-4. OpenAI said the company "works with more than 50 experts to provide early feedback in areas including AI safety and security."

As part of the announcement, OpenAI shared use cases from several companies that have tested GPT-4, including Duolingo, Stripe, Morgan Stanley, and the Government of Iceland. The company also shared a GPT-4 research blog, emphasizing that "GPT-4 still has many known limitations that we are working hard to address, such as social biases, hallucinations, and adversarial cues."

Computer Scientist and Brown University professor Suresh Venkatasubramanian remains cautiously optimistic about the news. In an interview with VentureBeat, he said: “I’d be interested to see how ChatGPT performs in a stress test under the public gaze, and I’m particularly interested in whether and how it can produce deceptively authentic content. , and what protections are in place to prevent this from happening.”

The above is the detailed content of OpenAI releases GPT-4, which has stronger question-solving ability. It also accepts image input, making it easy to understand memes!. For more information, please follow other related articles on the PHP Chinese website!