laitimes

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

author:New Hunan

Baidu will release the Chinese version of Chat GPT tomorrow, but the Chat GPT on the other side of the ocean has enlarged the move, and Chat GPT4 is coming. Fu Sheng, chairman and CEO of Cheetah Mobile, who has tried it, lamented that the Chat GPT4 ability is too terrifying, not only can understand text, but also read pictures, and the exam exceeds 90% of humans, and everyone must embrace this wave of technology.

The relevant discussion quickly rushed to the top of Weibo's hot search, and even hotter on overseas social media, almost everyone around the world is discussing it.

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

ChatGPT4 can read pictures and write essays, and the exam is more than 90% human

Just over four months before its release, after ChatGPT showed its amazing prowess, OpenAI dropped another nuclear bomb: GPT-4 was released.

In the blog post, OpenAI wrote: "We created GPT-4, the latest milestone in OpenAI's efforts to scale deep learning. GPT-4 is a large multimodal model (accepts image and text input, provides text output) that, while inferior to humans in many real-world scenarios, demonstrates human level on a variety of professional and academic benchmarks.

Greg Brockman, president and co-founder of OpenAI, shows the true power of GPT-4 – summarizing articles, writing code, filing taxes, writing poems... What GPT-3.5 can't do, GPT-4 easily takes.

For Xueba, the real opponent is coming, this time not humans, but robots.

OpenAI allowed GPT-4 to take many human general tests, and it turned out that it did perform significantly better on many tests and benchmarks than the previous generation: according to their test results, GPT-4's SAT score increased by 150 points, and now it can score 1410 out of 1600;

It can pass the mock bar exam with a score in the top 10% of test takers, compared to a GPT-3.5 score of around the bottom 10%;

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

ChatGPT4 test scores exceed ninety percent of humans

In both the SAT Reading Test and the SAT Math Test, GPT-4 scores can achieve leading rankings...

In addition, GPT-4 has made a qualitative leap - it can start processing images.

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

ChatGPT4 can read graphs

People who regularly use ChatGPT surely know that it can only process text, but GPT-4 began accepting images as an input medium.

In an example provided by OpenAI, GPT-4 accurately answers several questions about why network memes are funny.

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

GPT4 can recognize that this is a VGA interface and why this picture is funny (GPT4: The humor in this image comes from the ridiculous practice of plugging a large, outdated VGA connector into a small modern smartphone charging port.) )

However, the permission for image input has not yet been disclosed, so we have not seen more examples to prove GPT-4's image processing capabilities.

In addition, GPT-4 has begun to have a little sense of humor. It can already tell patterned, low-quality jokes — but at least it's beginning to understand the human trait of humor.

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

Greg Brockman, president and co-founder of OpenAI, personally demonstrated the true power of GPT-4 - summarizing articles, writing code, filing taxes, writing poems... What GPT-3.5 can't do, GPT-4 easily takes.

In the demo, Brockman used the "System" box to make GPT-4 successively "ChatGPT", "AI Programming Assistant", and "TaxGPT" to solve different problems.

In ChatGPT mode, GPT-4 can process more than 25,000 words of text, which can easily summarize the core content of a very long article, such as the summary points of OpenAI's eloquent article about GPT-4 released today.

It can even be distilled in a variety of strange forms, as in the demo—for example, summed up with words that all "start with G."

Or ask it to write these points into a poem.

In the "AI Programming Assistant" mode, it can also be easily written to generate a website, or more complexly, write a Discord-based robot, and if there is an error, such as calling a relatively new API and making an error, it does not even need to explain to it, but copies the error code into it, and it will automatically correct the error and generate new code.

Or turn it into TaxGPT, which calculates how much tax a couple has to pay based on tax laws, and also writes out the reasons for the calculation step by step so that people can review.

After trying it, Fu Sheng sighed: It is too good, it is the productivity change of the whole world

"chatGPT4 is very, very terrifying, it is a change in the productivity of the whole world, a wave of technology is coming, embrace it first!" Fu Sheng, chairman and CEO of Cheetah Mobile, lamented after the trial that everyone should care about the new AI, the new ChatGPT.

Hot search No. 1! Is the world really going to change? ChatGPT4 shocking release, ability "very scary"

Cheetah Mobile CEO Fu Sheng lamented that chatGPT4 is too terrifying

Although GPT-4 exploded the tech world with its power in just a few hours after its launch, its developers also pointed out some of its limitations. For example, although the accuracy of the answer has been significantly improved, it is still sometimes "fabricated facts", so it needs to be manually verified when using it.

Although GPT-4's capabilities have been greatly upgraded, the previous ChatGPT hallucinations and nonsense have not been completely changed. As can be seen from the previous official example, in order to maximize the capabilities of GPT-4, it is best to use the Chain-of-thought Prompt. It is also very simple to operate, just remind the AI to give steps when asking questions, which can greatly improve the accuracy of reasoning and calculation.

Of course, in addition to the performance aspect, there is also a noticeable difference that ChatGPT-4 is "bigger" than previous versions, which means that it has been trained on more data and is therefore more expensive to run. OpenAI only said it used Microsoft Azure to train the model, but did not publish details about the specific model size or the hardware used to train it.

This also means that the cost of using it is getting higher and higher. Unlike ChatGPT, the new model is not currently free to open beta, and while it encourages developers to sign up for a trial, it will need to be on a waiting list.

The new model, which will be offered to ChatGPT's paying user ChatGPT Plus ($20 per month), will also be available as part of an API that allows developers to pay to integrate AI into their apps. OpenAI says several companies have integrated GPT-4 into their products, including Duolingo, Stripe and Khan Academy.

Read on