00:02And congratulations on being shortlisted by the government, right?
00:05Building out the proprietary AI foundational models.
00:08Yeah.
00:09When do you expect a conclusion to that?
00:13And do you expect to win out here?
00:14It's not a single year conclusion.
00:16Every year we are going to get evaluated.
00:18It's kind of a tournament, that, like, survival game.
00:23Okay.
00:24Well, don't we love one of those games?
00:26But what do you think is your competitive edge here?
00:30So we have been built this LLM for almost four years, starting from the scratch.
00:36And we have a lot of exploration and experience and the technological advance.
00:41So the goal is to achieve the first place.
00:44Okay.
00:44Yeah.
00:45What the government and LG is trying to do is an open source model?
00:50So we try to build some sort of the open source model to facilitate all the other industry.
00:57But on top of that, we can also make the domain-specific model for the LG company as well.
01:03So it's a two-way.
01:05Why does the South Korean government think it's important?
01:07And why does LG work in this effort?
01:11So the South Korean new government really think the AI transformation is the key for the next stage, for all the industry.
01:21And there has been some major players in the Korean industry.
01:25And then they try to invite five major industries to compete with each other and then build some commercial usable LLM and provide that to all small industry.
01:37And doing that, we can get a lot of experience and evolution of the model, eventually facilitate the entire industry.
01:44It's really been interesting to see South Korea trying to accelerate this sector as a new growth engine for the nation.
01:52Is there a concern out there that this could be a bubble, that eventually you wouldn't need to build out so much infrastructure and so much investment into the sector?
02:02So we have talked about or the concern this bubble historically and many technological advancement.
02:10But for the AI, it has been there more than 30 years.
02:14And the new paradigm of the learning, learning from the data or the data-driven model has been there more than two decades.
02:21And we have constantly seen the progress.
02:24The Gen AI is just a result of that constant progress.
02:27It's not just some disruption, something to the scene, which has never been there.
02:35So it's very unlikely AI is a bubble.
02:38And we just constantly grow.
02:40And it tries to transform a lot of industry to the human life, daily passion.
02:46And LG, of course, the conglomerate itself in so many different sectors of business.
02:50How is the company going to enhance its own business and its own products through artificial intelligence?
02:57As everybody knows, LG has a lot of different affiliates, including electronics, energy solution, or the telecom service with the U+, or even like chemistry.
03:09So the way that we saw the evolution of the model, it first touches daily conversation, like changing or substituting the search engine, mostly for just Q&A.
03:21Later, we saw that there's a tremendous capability hidden there to touch the domain-specific stuff.
03:30And by introducing AI-agentic workflow that matches with industry, the workflow, we finally get a lot of transformational power.
03:40We are actually on that stage, starting from 2025 to 2026.
03:46Finally, just beyond the chatting machine, we are going to be able to see the AI module actually embedded into the industry, manufacture lines, and orchestrate a lot of jobs.
04:00Is that transformational potential when it comes to vertical AI and the integration into the business?
04:06Yes.
04:07When you talk about very soon and how that's going to be integrated into our daily lives, what's the timeline?
04:14So the timeline will be depending on the industry-specific data, how much data is already, and how much the experience and users are ready to do.
04:26So currently, in 2025, every agenting AI workflow started from just the user and the daily thing, like outlook, email check, and how to do the Google spreadsheet and things like that.
04:40But eventually, we are going to reach an industry-specific workflow.
04:44So in some sense, the workflow may be even more simple in the industry because it has to follow certain protocol, whereas the data and getting those data in industry is difficult.
04:58What a power to the LG is.
05:00LG really, they collaborate a lot between the subsidiaries and different affiliates.
05:06We are able to get the data, and we are able to tune the model really specific to that.
05:12And you're also collaborating with other companies.
05:14I understand Furiosa, you have a partnership there.
05:17Is that correct?
05:18We have a partnership for the AI chip industry.
05:20The Renegade chip.
05:21Yeah.
05:21What's different about that, and how does that apply to your own projects?
05:27XL1, the LLM, is what you're pursuing, right?
05:30So to train the model and to use the model, we need a lot of GPU resources.
05:37And the other name of the GPU is AI chip because it's basically the brain for the AI.
05:43But it happens depending on the external AI chips developed outside of the country.
05:50But over time, with this collaboration, we are seeing there's a lot of potentials, and there's even more benefit of using our domestic chips.
06:00So we are testing a lot, and over time, we are going to employ more of the domestic chips under our collaborations.
06:08Is circular deals and financing something that you think about?
06:13Because when we talk about a potential AI bubble and all of this money going between companies back and forth,
06:20there's a concern that this is how the bubble pops.
06:25Being in the field and as a researcher, is this something that you think about when you need the research funding to back your projects?
06:34Yeah, so I think there are two different aspects about that.
06:40Just in terms of the money flow, there has been a report from MIT that a lot of money has been invested to the major companies,
06:52but only 5% of that actually deploy that.
06:56And so far, there's no profit and loss expressions, even though there are a lot of investment.
07:02And I guess if there's no profit and loss yet, but at least in the research industry and the model performance,
07:12we saw that drastic performance boost so that everybody can use AI quickly in an easy fashion.
07:20So if nothing has been outputted, I probably have a big concern.
07:24But I can see like almost daily basis with big outputs from the industry.
07:30So I don't think that that's going to be simply a bubble.
07:35And then, yeah.
07:37The productivity boost and the efficiency boost, for example, on the other side of the coin is also an issue when you're thinking about jobs
07:43and how that could cause more layoffs instead of just enhancing people's lives.
07:49How do you assess that?
07:51I recall when I used to be a university student, there has been transition from a film camera to the digital camera.
08:00And I saw a lot of stories about film printing, film developing. They closed.
08:05Eventually, within three to five years, I saw way more order of magnitude more jobs has been created based on that technology.
08:16I thought similar stuff would come if we make or prepare everything ready on the right fashion, considering also risks, not only the capabilities.
Be the first to comment