In addition to Weibo, there is also WeChat
Please pay attention
WeChat public account
Shulou
2025-01-18 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > IT Information >
Share
Shulou(Shulou.com)11/24 Report--
CTOnews.com, August 18, as the AI model becomes more and more bloated, some research teams are advocating more efficient and smaller specialized models to replace the "large LLM language model."
According to Nature, OpenAI's ChatGPT is a new favorite in the technology world, but when ChatGPT encounters some mathematical problems that need to be deduced, it often makes mistakes. Nature magazine also cited an algebra problem to show the limitations of ChatGPT:
A straight line is parallel to y=4x+6 and passes through (5, 10). What is the ordinate of its intersection with the y-axis?
Although ChatGPT can sometimes give the right answer, the model still has a very high probability of getting the wrong answer, according to Nature. In an early derivation test conducted by the magazine, ChatGPT answered only 26% of the math problem sets at the American high school level.
Nature also said that in June 2022, a special mathematical computing model called Minerva broke through the industry's view that "big language models are not good at math". It is reported that Minerva answered 50% correctly in the math problem set, which surprised researchers in the AI field.
▲ image source "Nature" magazine
▲ image source "Nature" magazine
▲ image source "Nature" magazine
▲ source "Nature" magazine Microsoft Research machine learning expert S é bastien Bubeck said at the time: "in the AI circle, this is really a novelty."
The journal Nature claims that the results of the Minerva model support what the research team has long suspected: training larger AI models is not the right way to solve "logical problems".
However, Nature magazine also has some reservations about this view. They note that some research teams' examples show that "larger models only answer questions that happen to be related to the training set better than the ability to answer entirely new questions" (CTOnews.com Note: because there is too much training data in the model, similar questions happen to be compared in the training data, so they can give standard answers).
Nature also offers a series of tables showing the parameter comparison of today's large AI models:
▲ image source "Nature" magazine
▲ image source "Nature" magazine
▲ image source "Nature" magazine
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.
Views: 0
*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.
Continue with the installation of the previous hadoop.First, install zookooper1. Decompress zookoope
"Every 5-10 years, there's a rare product, a really special, very unusual product that's the most un
© 2024 shulou.com SLNews company. All rights reserved.