What language models are good at

If you read papers related to various language models, cases are usually divided into 7 categories. Most of the methods we use are derivatives of those large categories. If you go into more, there are problems that require more inference, such as solving math problems. The 7 listed below are basically areas that language models are good at, and other areas still require training. You can think of them as such.

Text Summarization

Information Extraction

Question Answering

Text Classification

Conversation

Coding Code Generation (a derivative of text generation)

Reasoning

Likewise, humans do well only when they are told to do something they are good at. Just as it would be very sad if you asked them to do something they are not good at and asked why they are not good at it, even if artificial intelligence cannot express sorrow yet, it would be somewhat difficult for us to get what we want. Therefore, I need to think about whether what I am ordering artificial intelligence to do now is really what the language model is good at.

Since language models are not search engines, many people approach them as search engines, but this is actually the biggest problem. While search engines look for existing content and display it, language models reassemble scattered content and create it on the fly. (Do you know why people are confused when you ask about the weather or the news?) Even for Bing, which is currently in service, if you look at how it works, information is extracted using a language model and searched using intent.

Propmt : 현재 2030년 액스포를 대한민국의 부산, 사우디아라비아의 리야드, 이탈리아의 로마가
경쟁을 하고 있다고 하는데 어느 곳이 개최할 확률이 가장 높나요?

Currently, Busan, South Korea, Riyadh, Saudi Arabia, and Rome, Italy are competing to host the 2030 World Expo. However, Riyadh, the capital of Saudi Arabia, was ultimately selected as the host of the 2030 World Expo. This decision was made after more than 160 countries participated in the vote, and Riyadh won the bid by winning 119 votes, beating out Rome and Busan.

This is how it works. You can see that it extracted the keyword " 2030 Expo Host City Competition ", searched it on a search engine, and summarized the results. If we put it in a difficult way, we can infer intent... or something like that, but you can understand it as understanding it perfectly. If we know this, we can now easily use artificial intelligence to draw pictures, summarize information, create slides, and create reports.

It may be used for commercial purposes with permission from the copyright holder, provided the source is cited.

Made with SlashPage