ChatGPT-4 is released! Improved accuracy, able to beat 90% of humans on the SAT

News on March 15th, on Tuesday, local time in the United States, the artificial intelligence research company OpenAI released its next-generation large-scale language model GPT-4, which is its new model that supports ChatGPT and Xinbi The latest AI large-scale language model for applications such as The company said the model performed beyond "human levels" in a number of professional tests.

OpenAI claims that ChatGPT-4 is larger than the pre-iteration GPT-3.5, which means it has received more data training and has more weights (parameters) in the model file, which also makes it operating costs are higher. The company claims that the model is "more creative and collaborative than ever" and "can solve difficult problems more accurately." It can parse text and image input, although it can only respond with text.

Many researchers in the field currently believe that many of the recent advances in AI have come from running ever-larger models on thousands of supercomputers, a training process that can cost tens of millions of dollars. GPT-4 is an example of the focus on “scaling up” to achieve better results.

OpenAI admitted that the company used Azure, Microsoft’s cloud computing platform, to train its models. Microsoft has invested billions of dollars in OpenAI. Citing competitive reasons, OpenAI did not release details such as the exact size of the model or the hardware used to train it, which could be used to reconstruct the model.OpenAI’s GPT large language model powers many of the AI demos that have wowed people in the tech industry over the past six months, including Bing’s AI chatbot

and ChatGPT.

ChatGPT-4 is a preview of the latest advances in language models that may start trickling down to consumer products like chatbots in the coming weeks. Microsoft said on Tuesday that Bing's AI chatbot uses the GPT-4 model.

OpenAI claims that the new model will produce fewer factually incorrect answers, go off topic less often, talk less about forbidden topics, and even perform better than humans on many standardized tests.

For example, the company said that GPT-4 ranked in the top 10% of all candidates in the mock bar exam, in the top 7% in the SAT reading test, and in the top 11 in the SAT math test. %.

However, OpenAI warns that the new model is not perfect yet and in many cases is less capable than humans. For example, GPT-4 still suffers from so-called “hallucinations” or fabricated stories and is factually unreliable. When it makes a mistake, it still tends to insist that it is right. OpenAI CEO Sam Altman tweeted that GPT-4 "still has flaws and big limitations," but that "it still impresses you the first time you use it." ."

OpenAI said in a blog post: "GPT-4 still has many known limitations that we are working hard to address, such as social bias, hallucinations, and hostile replies. In a casual conversation, The difference between GPT-3.5 and GPT-4 is small. And when the complexity of the task reaches a sufficient threshold, the difference becomes apparent: GPT-4 is more reliable, more creative, and capable than GPT-3.5 More nuanced instructions."

OpenAI said it has worked with multiple companies to integrate GPT-4 into their products, including Duolingo, Stripe, and Khan Academy, among others. The new model is available to users through OpenAI’s $20-a-month ChatGPT subscription service, ChatGPT Plus, and powers Microsoft’s Bing chatbot. At the same time, GPT-4 will also be available as part of an API that allows programmers to integrate AI into their own applications.

The above is the detailed content of ChatGPT-4 is released! Improved accuracy, able to beat 90% of humans on the SAT. For more information, please follow other related articles on the PHP Chinese website!