Shulou(Shulou.com)11/24 Report--

OpenAI, the US artificial intelligence research laboratory, released the latest GPT-4 language model for its popular chat robot ChatGPT on Tuesday, just four months after the launch of ChatGPT.

Foreign media said that compared with the GPT-3.5 model originally used by ChatGPT, GPT-4 has brought ten improvements and improved accuracy and other functions, but there are still some shortcomings such as errors and "hallucinations".

The following are the main improvements and deficiencies of GPT-4 summarized by foreign media:

Chris Nicholson (Chris Nicholson), a partner at venture capital firm Page One Ventures and an expert in more precise artificial intelligence, told GPT-4 that he was native English and did not speak Spanish. He wants GPT-4 to give him a syllabus that will teach him the basics of Spain. As a result, GPT-4 provides a detailed and structured syllabus. It even provides a wide range of skills for learning and memorizing Spanish words, although not all suggestions are pertinent.

Improvement of accuracy

GPT-4 is more accurate than GPT-3.5. It also improves accuracy. When Oren Etzioni, an artificial intelligence researcher and professor, first tried to use GPT-4, he asked a straightforward question: "what is the relationship between Oren Ezioni and Eli Ezione (Eli Etzioni)?" The robot's response is correct, saying that the two are father-son relationship. GPT-3.5, by contrast, gave the wrong answer, calling the two brothers. However, there are also errors in GPT-4. It says Oren is CEO of the Oren Institute of artificial Intelligence, but in fact Oren has resigned from the position.

Describe the image in more detail

Can describe images in detail GPT-4 can make surprisingly detailed descriptions of images and text. Greg Brockmann, president and co-founder of OpenAI, demonstrated how the system details images from the Hubble Space Telescope. It can also answer questions about images. If you give a picture of the inside of the refrigerator, it will advise you to cook a few meals with what you have at hand.

OpenAI, which specializes in standardized tests, says the new system can achieve the top 10 per cent in the uniform Bar qualification examination (UBE) in 41 states and territories in the US. According to the company's test, it can also score 1300 out of 1600 on the SAT and 5 out of 5 on the college biology, calculus, macroeconomics, psychology, statistics and history prerequisite exam for high school students.

Not good at talking about the future although the new robot seems to be able to reason about what has happened, it is not so good at it when asked to make assumptions about the future. It seems to be borrowing from other people's words, rather than creating new guesses.

Still hallucinate. New robots still make things up. This problem, known as the artificial intelligence "hallucination", haunts all leading chatbots. Because the system does not know what is true and what is false, they may generate completely wrong text. When asked to provide the address of a website describing the latest cancer research, it sometimes generates an Internet address that does not exist.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.