Each Holds a Unique Philosophy in AI Development

Kazuki Kyakuno (Hereafter, Kyakuno): It’s been two years since the launch of “2WINS.” Are there any philosophies or consistent thoughts in development that have not changed?

Ryuto Ogawa (Hereafter, Ogawa): The idea of creating something that no one has ever made, or thought could not be made, and delivering it to society or a part of it, has not changed. I believe I will pursue this for my whole life. As a result, I want to become known as someone who created “those things.” That is the central belief guiding our development. We want to create something that even we as the creators feel is useful, and build a team that can have a social impact. I see AI as one tool to achieve this.

Kyakuno: That’s wonderful. How do you think the AI your team develops will change people’s lives in the future?

Ogawa: As AI becomes more widespread in people’s lives, I believe it will allow people to focus more on the essential aspects of their time. In today’s world, before you can start doing what you really want, there are many other things you must do first. However, I think we can provide a world where you can spend your time and money on the essentials without doing those other things. As a result, I hope we can become an entity that delivers the kind of social impact I mentioned earlier. To do this, we aim to start developing solutions within our current reach and expand that circle. Could you share your thoughts, Mr. Kyakuno?

Kyakuno: Personally, the happiest moment for me is when I use an app I created myself. As I use what I’ve created every day, its precision improves, and I feel a sense of fulfillment. Since I designed it myself, it naturally has the best UI/UX and APIs, and there’s no stress in using it. Becoming a user myself and delivering products and services that satisfy our customers is a key point I cherish.

Ogawa: In that sense, perhaps “personalization” will become one of the keywords for apps in the future.

Kyakuno: Additionally, the attitude of “creating quality from quantity” while steadily progressing in development has also always been the same. We have created various things in this way, but currently, the most frequently used is our own “ailia SDK (Ailia SDK, a world-class inference engine developed by Axell).” I use the ailia SDK’s APIs to build apps daily, translate instructions for our engineers within the company into English using “ailia DX Insight (Axell’s AI DX app),” and am even using AI to create the transcript for this interview right now.

Ogawa: The more you use it, the more you realize its unexpected potential, which is one of the interesting aspects of AI.

Kyakuno: AI is evolving rapidly, and just as the internet has empowered people’s lives, the acceleration of AI will also exponentially increase the potential each individual can bring to their work.

Ogawa: Indeed, the proliferation of the internet was likely a turning point similar to today. However, I also think that the areas changed by the internet are surprisingly narrow. It brought significant changes to advertising and e-commerce, but there were many areas it did not change. In the case of AI, its influence is much larger and is expected to bring change to every sector. That is something I look forward to.

Kyakuno: Even looking at our customers who have implemented “ailia SDK,” as Mr. Ogawa mentioned, they come from a variety of industries. As creators, we constantly feel the “breadth” of AI.

The Current AI Scene, Where Developers Are Also Overwhelmed by Its Speed

Ogawa: Since AI is our specialty in business, of course, we can’t afford to fall behind on the latest AI trends. But to be honest, the flow of information is too fast, and it’s impossible for me alone to keep up (laughs). Even my colleagues, who are better at gathering information than I am, say that “it’s tough to catch all the information,” as new AI-related news flies around at a dizzying pace every day.

Kyakuno: New papers on AI are indeed published every day.

Ogawa: Exactly. I might read some of the latest papers and think, “This could bring novelty to our business!” But before you know it, the next piece of information becomes the talk of the town. Those who know about AI probably feel the speed of this information the most.

Kyakuno: Recently, we’ve seen more initiatives that cross over with other fields, like AI × robotics.

Ogawa: There are many functions where AI is taking the lead, like new technologies in infrastructure such as autonomous driving, or automated harvesting of fruit in agriculture. I feel the “atmosphere” of potential expansion is still very much alive. Watching all this, I realize again how incredibly fast the AI scene is moving.

Kyakuno: While the evolution of AI is fast, there are still cases where technologies like “YOLOv3” from around 2018 are still in use. This is because there was a period from 2020 to 2022, before “ChatGPT” emerged, when the global AI scene temporarily slowed down, with many companies pausing their research. With the recent advancements in AI, these activities have restarted. Thus, when integrating the latest AI into DX (Digital Transformation), using some SDK (Software Development Kit) makes it easier to catch up, which is why there’s a demand for our “ailia SDK” or embedded AI. Moreover, since 2023, instead of CNNs (Convolutional Neural Networks), Vision Transformers have become mainstream, further enhancing the evaluation of “ailia SDK.”

Ogawa: By 2022, “Stable Diffusion” appeared in the generative AI scene, and in two years, it became possible to generate not only still images but also videos.

Kyakuno: The sense of speed is fast, and everything is non-continuous. It’s almost abnormal how new architectures keep emerging every week. Until the latter half of 2023, new architectures continued to appear, but now the AI scene seems to have shifted to a “brute force” approach, processing more data with the same architectures. However, if you let your guard down, incredible new developments will spring up, which is both the interesting and vigilant part of the AI scene (laughs).

Ogawa: In development, there seems to be a divergence into two paths as Mr. Kyakuno mentioned: the “brute force AI” that advances with large computing power, and “constrained AI” used in areas where high computational demand is not feasible or only needed locally. I’ve heard there are also movements to combine many small AIs to counter the robust large AI models. Mr. Kyakuno, how do you see this evolving?

Kyakuno: There is a benefit that the “large AI models” at the top are becoming stronger, reducing the labor needed to create “small AI models.” For example, if you wanted an AI to recognize an ‘umbrella’ placed in a conference room, you previously had to take photos of umbrellas and annotate them yourself. But now, just collecting images can allow the base model like Detic (an image recognition model developed by Meta) to do all the annotation automatically, and all that’s left is to run training for YOLOX (a real-time object detection model), which greatly simplifies the process. If the use case for AI is defined, I believe development will become even more accessible in the future.

Ogawa: About two years ago, I manually annotated data for cars and people for autonomous driving with YOLO, and now I think, “I wouldn’t have needed to do that work if it were today…” which gives me mixed feelings (laughs).

Kyakuno: It’s really helpful for us in development when such convenient AI models are made public (laughs).

Link

2WINS Inc.
https://www.2wins.ai

For inquiries about AI development, products, and services, please contact:
https://ailia.ai