To organize the articles I've written for my blog, I leveraged generative AI (Gemini) to create my own website.
katoshi's Research Notes https://katoshi-mfacet.github.io/
This site is automatically generated from my original blog article drafts, written in Japanese.
Its features include:
- Automatic generation from article drafts
- Article organization by categorization and tagging
- Support for 30 languages and accessibility
Basic Mechanism
The basic mechanism involves a self-made program based on the Astro framework, which automatically generates HTML files from article drafts.
I even created this program itself by chatting with Google's Gemini.
Thanks to this mechanism, once an article draft is written and the regeneration process is executed, the HTML files are automatically updated and reflected on this website.
Categorization and Tagging
I have also developed a separate program for categorization and tagging.
This program passes articles to Gemini via an API to automatically categorize and tag them.
When provided with a list of categories and tags along with the article, Gemini interprets the article's meaning and skillfully suggests appropriate ones.
Furthermore, the category and tag lists themselves are determined by another custom program that extracts them from past articles. Here, too, I leverage Gemini.
Past articles are sequentially passed to Gemini via an API, which then outputs candidate categories and tags. These extracted candidates from all articles are then fed back to Gemini to finalize the comprehensive category and tag lists.
This entire process is also automated by the program.
Multilingual Translation
Translation is essential for multilingualization. Naturally, Gemini is utilized for this translation as well.
There are two patterns for translation:
One is the translation of common strings within the website, independent of specific articles, such as menu item names and self-introductions.
The other is the translation of the article drafts themselves.
For both of these, I created a custom program that uses Gemini's API to perform the translations.
Accessibility
Considering users with visual impairments who listen to article content via audio, and those with difficulty using a mouse who navigate websites solely with keyboard operations, incorporating several enhancements into the HTML files improves accessibility.
Regarding accessibility, I had very little knowledge; Gemini actually suggested these improvements during our programming chat.
And for these HTML modifications to enhance accessibility, I also learned how to implement them by chatting with Gemini.
Disappearance of Walls
Generative AI was utilized in various ways for the creation of this website, including program development, natural language processing for translation and organizing categories and tags, and even suggesting subtle points like accessibility that I might have overlooked.
Furthermore, by creating a mechanism for automatic updates upon adding articles, including HTML generation and natural language processing for categories and tags, I've been able to make this website one that will continue to grow with each new article.
Through the creation of this website, I truly felt how easily various barriers can be overcome with generative AI.
First and foremost is the language barrier. Supporting 30 languages would have been utterly impossible for an individual traditionally, even considering translation.
Additionally, there are concerns about whether translated blogs convey the intended nuance and if the expressions are unnatural or offensive to native speakers.
Generative AI's translations can convey nuances more accurately and use more natural expressions than traditional machine translation. Moreover, the translated output can be re-entered into the generative AI to check for unnatural or inappropriate phrasing.
From the perspective of website multilingualization, properly handling elements that vary in expression by language, such as dates and units, was another difficult point.
For example, if there are 1, 2, and 10 articles in three respective categories, in Japanese, it's sufficient to simply add the unit "記事" (kiji - article/items) after the count, like "1記事" (1 article), "2記事" (2 articles), "10記事" (10 articles).
However, in English, it's necessary to distinguish between singular and plural forms, such as "1 article," "2 articles," "10 articles." Furthermore, in some languages, expressions can reportedly change even for small versus large plural counts.
Moreover, for languages like Arabic that are written from right to left, the entire website layout must be made natural, following the reader's line of sight from right to left. If arrows are used in text or images, the necessity of horizontally flipping them must also be considered. These points are also addressed by having generative AI check them.
By working on website multilingualization with generative AI, I was able to meticulously address details that I would have overlooked and couldn't have accounted for with traditional methods.
The same applies to accessibility considerations. Previously, I could only provide considerations for people who could view the website in the same way I could.
However, generative AI easily incorporates considerations that I wouldn't notice, or that I might hesitate to address due to the effort involved.
While multilingualization and accessibility may not be perfect yet, I believe the quality is overwhelmingly higher than what I could have achieved by thinking and researching on my own.
In this way, generative AI has eliminated many barriers to the endeavor of disseminating information through blog articles.
Conclusion
I am a system engineer with extensive programming experience. While I don't create websites for work, I've made several homepages as a hobby in the past.
Leveraging this experience and my interactions with generative AI, I was able to build the automated generation system for this multilingual blog site in about two weeks.
Without generative AI, I would never have even considered multilingual support in the first place. In that sense, it can be said that I overcame the barrier of imagination.
Furthermore, considering the effort involved in categorizing and tagging articles every time they are added, it's highly likely I would have stopped updating the site after its initial creation. With the automation provided by generative AI's natural language processing, I was able to overcome the maintenance and update barriers as well.
Moreover, this system can be built by individuals without programming or website creation experience, like myself. If you were to feed this article to a generative AI like Gemini and express your desire to build something similar, it should be able to guide you through the process.
While I could publish the program I created for widespread use, given that generative AI is becoming a full-fledged software engineer, the most valuable information to share now is likely the explanation of the ideas and mechanisms, as presented in this article, rather than the program itself. Ideas and fundamental mechanisms are even easier to modify, enhance, or combine than programs.
This suggests that while the barriers to software development and website creation are disappearing, so too are the barriers to individual information dissemination.
Technologically, the internet has virtually eliminated the barriers to information exchange, yet we are still hindered by barriers such such as language and accessibility.
While recipients can overcome some barriers to a certain extent through their own efforts, such as machine translation and text-to-speech, there are also parts that cannot be overcome unless the information sender provides support and consideration.
It is precisely these barriers that information senders must overcome that generative AI eliminates.
Even if the barriers of language and accessibility disappear, there will undoubtedly be further barriers such as differences in culture, customs, and values. These may be even more difficult to overcome.
However, to overcome these difficult barriers, we must first overcome the ones that lie before them. Once we reach the foot of those walls, new ideas and techniques for scaling them will surely emerge.
It may be that we are entering an era where walls are disappearing from the world. Through the creation of this website, that is precisely what I felt.