日本語

ELI5

Explain it like I’m five
Ask, and itmakes somethingnew.

生成AI(せいせいエーアイ)What is generative AI?

日本語English

In one lineGenerative AI is artificial intelligence that creates new content, such as text, images, audio and video, based on patterns learned from large amounts of data.

Generative AI is AI that makes new things, like writing, pictures, music and video. It learns patterns in words and pictures from huge amounts of data. Then it puts together something that fits your request. It became part of daily life after ChatGPT, a chat AI, came out in 2022. People use it to draft emails, brainstorm ideas, translate and more. But it can also write mistakes with total confidence. So checking is still a human job. It is a good partner. Handing it everything is still too soon. It has that in common with interns.

Common mix-upIt does not just copy text from the internet. It builds new sentences from the patterns it has learned.
Words to know / 単語メモ
pattern
パターン
request
指示
draft
下書きする
confidence
自信
translate
翻訳する

Go deeper

How it works: guessing the next word

Most text-writing generative AI is a “large language model.” It learns from huge amounts of text which words tend to come next, then builds sentences by predicting one word at a time. A design called the Transformer, published in 2017, let these models handle how all the words in a long passage relate to each other, and it became the base of today’s generative AI.

Mistakes and copyright

Because it picks the most likely-sounding next words, generative AI sometimes writes false things as if they were true. This is called a “hallucination.” And if what it makes looks very much like an existing work, copyright can become an issue. In March 2024, Japan’s Agency for Cultural Affairs published its General Understanding on AI and Copyright, treating training and the use of generated output separately.

How people use it
  • “I drafted the proposal with generative AI, then checked every number myself.”
  • “I asked the AI for its source, and it gave me a book that doesn’t exist.”
Questions people ask
When was ChatGPT released?
OpenAI, an American company, released it on November 30, 2022. That is when generative AI suddenly became part of everyday life.
Are its answers correct?
Often, but not always. It can state wrong things confidently, so check numbers, names and sources against the original.
What should I watch out for at work?
Do not type in company secrets or personal information, and check the output before you use it. If your workplace has rules for AI, follow them.

Sources:Wikipedia「生成的人工知能」、arXiv「Attention Is All You Need」(2017年)、文化庁「AIと著作権について」、Wikipedia「ChatGPT」

Your turn

Do you use generative AI?

Tap to see what everyone said.

SharePostLINE

Related words

Source (in Japanese): Wikipediaで「生成AI」を調べる