note’s New Challenge! Will Rewards for AI Training Come to Life?
Introduction
Hello, I’m Easygoing.
This time, I was really intrigued by the recently announced initiative to set compensation for AI training, so I decided to look into it myself.
AI Trains on Web Information for Free
I’ve previously written an article about AI and copyright.
In Japan, AI training on copyrighted works for free is permitted under an exception in copyright law.
Article 30, Paragraph 4 of the Copyright Act
A copyrighted work may be used by any means deemed necessary, regardless of the method, in the following cases or other cases where the purpose is not to personally enjoy or allow others to enjoy the thoughts or feelings expressed in the work, provided it does not unreasonably harm the interests of the copyright holder in light of the type and use of the work and the manner of its exploitation:
(Omitted)
When used for information analysis (meaning the extraction, comparison, classification, or other analysis of information such as language, sound, or images constituting a large number of works or other vast amounts of data; the same applies in Article 47-5, Paragraph 1, Item 2).
In addition to the cases listed in the preceding two items, when used in computer information processing without human perception of the work’s expression or for other uses (excluding the execution of a program in the case of a copyrighted program).
| Model Name | Release Date | Developer | Country |
|---|---|---|---|
| ChatGPT | November 2022 | OpenAI | USA |
| Claude | December 2022 | Anthropic | USA |
| Llama | February 2023 | Meta | USA |
| Mistral | September 2023 | Mistral AI | France |
| Qwen | August 2023 | Alibaba | China |
| DeepSeek | October 2023 | DeepSeek AI | China |
| Grok | November 2023 | xAI | USA |
| Gemini | December 2023 | USA | |
| Phi | December 2023 | Microsoft | USA |
Currently, AI companies like Google (USA), OpenAI (USA), and Alibaba (China) are collecting data from the Japanese web without violating existing laws.
Using this collected data, AI is trained to create conversational models like ChatGPT, Gemini, and Qwen, which can understand Japanese at a high level, not just English or Chinese.
Since AI is currently training legally, will a system to demand compensation for it actually work?
“Indicating Refusal for Generative AI Training”
On February 17, 2025, note implemented an intriguing feature.
That feature is the ability to indicate refusal for generative AI training.
note’s Article Settings Screen
Today, AI is widely used in daily life for things like web searches and translations, but as AI advances, the idea of not wanting one’s copyrighted works used for training has also gained traction.
I think the “indicating refusal for generative AI training” feature was introduced in response to this sentiment, but what does it actually mean in practice?
The Agency of Cultural Affairs Recommends Blocking with robots.txt
The Agency of Cultural Affairs recommends using robots.txt to block crawlers—unmanned programs that collect information from the internet—if you don’t want your copyrighted works used for AI training.
- Crawler: An unmanned program that collects information from the internet.
- robots.txt: A site owner’s declaration of intent to crawlers.
2024 Copyright Seminar “AI and Copyright II” 44:43~
Since indicating intent via robots.txt is a globally recognized standard, setting it according to the Agency’s guidelines allows you to communicate your wishes to crawlers worldwide.
note’s robots.txt Doesn’t Block AI Training
The robots.txt file, which declares intent to crawlers, is set for an entire site.
You can check note’s robots.txt at the following URL:
Contents of note’s robots.txt
- Permits crawling of the entire site.
- Prohibits access to unpublished pages, individual user pages, etc.
note’s robots.txt allows major crawlers like Google, Bing, and Twitter to freely crawl the site, except for unpublished pages and personal user pages.
note’s robots.txt doesn’t impose specific restrictions on crawlers collecting AI training data, such as GPTbot or Google-Extended.
Even if you enable the “indicating refusal for generative AI training” option, it doesn’t change the robots.txt entry for individual creators.
So, What Does It Mean?
So, what’s the point of the “indicating refusal for generative AI training” option?
I created a test account on note and activated this feature to compare page content, but no clear differences were observed.
I think the “indicating refusal for generative AI training” feature is still in the preparation phase, and for now, they’re gauging how many users sign up for it.
Recent Approach:
Recently, the use of the tag has been spreading as a way to indicate refusal for generative AI training, distinct from robots.txt.
The tag is a method to declare refusal of AI training at the top of a web page, and unlike robots.txt, it can be specified on a per-page basis.
I predict that note’s “indicating refusal for generative AI training” feature will align with this approach.
However, the tag is a weaker declaration than robots.txt, and at present, no AI companies have stated they will comply with it.
Even if you indicate refusal for generative AI training, it’s likely very difficult to actually prevent AI from training on your content.
What Is note Aiming For?
note is a company that actively adopts new technologies, so I don’t think they’d implement a feature with little effect.
So, what is note aiming for?
note Is Japan’s Largest Blog Site
As of March 2025, note is the most visited blog site in Japan.
noteが、国内ブログサービスでアクセス数トップになりました🎊
国内ウェブサイト全体のアクセス数ランキングでも14位。クリエイターのみなさん、いつもありがとうございます！https://t.co/Bzu9KumMBz pic.twitter.com/sZsalYK4pH— note (@note_PR) August 6, 2024
Since the 2010s, communication tools have shifted from blogs to SNS, and many major blog sites that respected veteran big-name bloggers and failed to enact fundamental reforms have faded away.
In that context, note invested in cutting-edge systems despite operating at a loss, creating a lightweight blog site that others couldn’t replicate.
Even articles written by unknown writers on note outrank major websites and appear high in search results starting the day after publication.
This aggressive stance by note has earned widespread support, leading to a surge in new users.
note Keeps Pushing Forward!
In December 2022, note went public despite operating at a loss, and in May 2024, it finally turned a profit.
Amid this, surprising news was announced.
In January 2025, note announced it would accept capital from Google, causing its stock price to soar to roughly double its pre-announcement value.
Why Partner with Google Now?
Why did note choose this timing to partner with Google?
As mentioned earlier, note achieved profitability on its own starting in 2024.
Unlike many other websites, note doesn’t use Google AdSense or rely on Google’s ads.
Since its founding, note has operated with minimal advertising, which was likely possible because it kept its distance from giant IT companies like Google.
Create, Connect, Deliver
Here’s a quote from note’s CEO, Sadaki Kato, about note’s philosophy:
Back then, the only way to monetize online content was almost exclusively through advertising. Advertising itself is neutral and not inherently bad, but we thought a situation where it was the only option was problematic.
There are two reasons for this. First, when you rely on ad revenue, you want to maximize page views, which tends to lead to escalating sensationalism. This ties into current issues like defamation and online outrage. Second, copy-pasting and plagiarism. These are common online problems, but they reduce the cost of content creation, which in turn boosts ad profitability.
To address these two internet challenges, we believed non-advertising monetization methods were necessary, so we built a payment feature into note from the start. Not everyone needs to use this feature, but we think it’s important to have it as an option.
Unless we change the cycle that forms the foundation of creation, the flow of creativity will dry up. That’s why this company has focused on monetization as well as creation from the beginning. The “continue” in note’s mission—“to enable everyone to start and keep creating”—includes monetization.
Sharing the Profits
When YouTube launched in Japan in 2007, it was initially flooded with copyright-infringing content, earning it a reputation as a piracy site.
Over time, YouTube introduced a profit-sharing system, which reduced piracy and helped it grow into a platform surpassing traditional media.
Similarly, Netflix, a video streaming service where users pay a low monthly fee to enjoy content easily, has gained a massive following since launching in Japan in 2015.
Netflix’s rise has helped curb piracy, showing that properly distributing profits enables everyone to keep creating.
Setting Compensation for AI Training
YouTube and Netflix originated in the U.S. and were introduced to Japan.
In contrast, setting compensation for AI training is a new endeavor that hasn’t been realized anywhere in the world.
There are opinions both for and against AI training on copyrighted works, and opposition remains strong.
If a profit-sharing system like the one that succeeded in video streaming works well, it could help resolve this tricky issue.
note’s New Challenge!
Today, Japanese content like manga and anime is globally acclaimed.
As Japan’s largest creator platform, note hosts a vast amount of content created by countless unknown creators who keep producing.
I think note is challenging itself to create new value for the works crafted by these creators.
Whether it’s setting compensation for AI training or indicating refusal for generative AI training, these are both powerful cards note can use as the voice of creators when negotiating with AI companies.
In an era where AI training on internet data for free is the norm, I believe note is taking a big risk by leveraging its content as a weapon.
Do you want to join this challenge?
Sign-ups are open until Thursday, March 6, 2025
Thank you for reading to the end!