CEO Coc Coc talks about the new game with AI: We are used to challenges!

Although admitting that the Coc Coc AI Chat product was launched when AI chatbots in general barely had a clear business model, CEO Nguyen Vu Anh still said that, for the benefit of users, this is a trend. that Coc Coc could not stand outside.

*CEO Nguyen Vu Anh is a French overseas Vietnamese

The launch of the ChatGPT product at the end of 2022 created an explosion in the technology industry and has since attracted a lot of interest, even for those who do not know much about technology. can use ChatGPT. Before this event, what are your thoughts? 

 In the history of technology products, ChatGPT within two months reached 100 million users, the fastest in history. To reach that number, Facebook took more than 4 years.

In my opinion, this breakthrough comes from the fact that the large language model (LLM), compared to previous content processing products, is extremely excellent.

In essence, LLM created a new user experience and user interface (UX/UI). As Microsoft points out, there are three user interface revolutions. The first revolution is when users access information with keyboard and mouse (PC), then touch screen (smartphone) and now chat. 

 This new interface could completely change the way users interact with content, which is extremely important for technology companies.

 How is the reception of Vietnamese users with AI Chat products?

In fact, ChatGPT has not officially operated in Vietnam (users cannot create an account and pay in Vietnam), so it can be seen that they have not focused on Vietnam. But as we tracked, there are also quite a few people using ChatGPT in Vietnam through Coc Coc browser.

 We see the market is quite new, and there are many opportunities, so we jump in, with two orientations: one is to focus on the Vietnamese market, the other is to integrate with search engines, for example ChatGPT if you ask for information. After 2021, there will be no answer.

 What orientation will Coc Coc have different from competitors when deploying its products?

In comparison, Coc Coc's product will be closer to Microsoft's Bing AI product, because they also use a large language model, integrated with search engines.

My guess is that, just guessing, chatbots and search engines will probably merge someday in the future. The starting point is to summarize the information of one page, then expand to summarize the information of many pages, and progress to the chatbot can answer the user's wishes much faster than the search engines. .

Currently, it can be seen that Bing AI is merging these two features, while we intentionally separate those two features into Coc Coc AI Chat and Coc Coc AI Search.

If you try Bing's AI chatbot, you will find it quite limited in producing highly creative content, as the mechanism is always search engine-based. But generative AI is also good in that it will go crazy at times. A chatbot based solely on search engines will be weak on creativity.

 User expectations of search engines should be accurate, while chat should be creative. To create an optimal tool is also very difficult.

 What is the difference when implementing this new product for the previous product?

 The product we've worked on before, a search engine, or a browser, isn't new, so we've got a clear picture of our users and business model, so we know how we can compete. . But chatbot AI is completely new and many users actually do not know how to take advantage of it effectively.

 So what is the possible business model for AI chatbot products based on LLM, according to you, what could be?

 If talking about the business model, even the big players are not clear, OpenAI opening ChatGPT Plus for 20 USD/month is just a test, still heavy loss. 

Due to the lack of a clear business model, the current revenue outlook is not optimistic. Investing in this area, for some giants like Microsoft, the goal is also said to beat Google's business model.

Like I said, LLM creates a new user interface. This new interface will likely replace the current main tools of the internet, the browser and the search engine, as these are the two basic tools for accessing content.

Is ChatGPT excellent? Very excellent. It is possible to reach 100 million users so quickly, because it interacts with users in a very natural way. Vietnamese may not be "smooth" but English is very good, very easy to apply, so the risk of being replaced by search engines in the future is quite obvious.

Therefore, many analysts also think that Google has not made this product because it will probably disrupt their own business model, they will have to consider because the search engine is the money making machine of the company. Surname. If not for that reason, perhaps it was Google who launched the LLM-based chat product first.

 With such resource consumption, the most likely scenario, in my opinion, is that there will be only a few companies that maintain the LLM model in the world, then they sell the basic service as LLM. As for using LLM to create products with its own business model, many companies can be creative.

 If you say, chatbot AI will probably disable the business model of search engines, will Coc Coc consider not continuing to develop AI chat?

No, you have to. If you look from the investor side, you will also see that AI is very hot (hot). If doing AI, nature will get their attention. Coc Coc is a B2C service provider, but there are not many AI products that can directly interact with end users.

 The current direction of Coc Coc is as a content search engine, interacting with content, but generative AI is fundamentally changing that, staying out of the game is unreasonable.

Compared to Google, Coc Coc is not a dominant player in the market, so it will be easier to create new ones, and be ready to disrupt the existing business model. So definitely have to play, and find your way. 

 What is the spirit of the Coc Coc team when developing such a very new and challenging product?

Engineers can just do something new. They like challenges (laughs).

 We are used to the challenge. Competing with Google, or changing business models, is equally challenging. We see this trend as another opportunity to compete with Google.

What will Coc Coc's approach when the product does not have a clear business model?

 Our approach is to launch the product first. The first goal is to learn how to apply this new technology. More importantly, Coc Coc wants to learn about users. When users try the product, Coc Coc will know how they use it, and how the product brings value to them, to find use cases. 

Some usecases are actually popular in other countries, for example, in the education sector, or the content creation industry, the software industry... Applications are using ChatGPT like Duolinguo (which integrates ChatGPT into the content). language learning), Snapchat… found their usecase. At this point, Coc Coc's strength is that it already has a browser product and a fairly large user base to distribute AI Chat tools.

In my opinion, an effective business model will probably still be advertising, but a reasonable mechanism must be found, and with that, the sale of accounts (subscriptions).

Currently, the effects of ChatGPT are still controversial when it can cause many jobs in society to be replaced, causing unemployment. Many places support this tool, while there are also places that prohibit it. When developing a similar tool, what does Coc Coc think about this?

Important is the legal framework of the government. China is ahead on this issue because it is very cautious, of course, in return, it will limit innovation to a certain extent.

 Risks like using AI to create fake news and manipulation are very dangerous. But in fact, without AI, social networks already have that risk. Technology has no way to go in the opposite direction, but will develop. It is important that governments have a legal framework in place to protect users.

With the goal of creating strong AI Chat in Vietnamese, will Coc Coc's orientation be fine tuned or run your own LLM?

Coc Coc alone is definitely not qualified to develop LLM like OpenAI, the current story is to find a usecase to fine tune and operate it yourself.

"Ask me anything" (Ask me anything) like Open AI is doing, is actually just the beginning, because it has not really brought daily value to the majority of users. The thing to do is to discover that and bring it to the user.

 As for the future, we actually have a wish, but we can't confirm what its future will be.

Obviously, it can be seen that all LLMs have the problem that Vietnamese is not good. For example, in OpenAI's trained data, only 1% is Vietnamese, the rest is mostly English and other languages. Vietnamese will be "low resource language", which means there are not many information resources.

In fact, this idea also has a limitation in that, OpenAI's LLM is excellent because it is trained in so many languages. If you only train in English, you can't be so good.

We are in discussions with some of our Vietnamese partners, working on AI, to see if there is an opportunity to build an LLM with data trained in English and Vietnamese, where the percentage of Vietnamese higher than existing LLMs or not.

The road is still long, and I don't expect the quality of the content to be equal, because the cost of running an LLM like OpenAI is a crazy number.

We know this is unlikely to be done by Coc Coc alone, so we will need to coordinate with our partners. Currently, there is an opensource project in the US, developed by overseas Vietnamese, Coc Coc side is participating to prepare the data. Vietnamese data is indeed limited, but we also have a strong point in the search engine platform. We will break down the usecases to do, for example, in each major.

With LLM, we look at two core issues: one is quality, the other is cost. Quality is definitely dependent on the amount of data used for training. If the input has a lot of data in Vietnamese, it is expected that the Vietnamese content will be better. On the other hand, training only two languages ​​will help lower costs. If we can come up with a better model for processing Vietnamese, more useful for Vietnamese people, it should definitely be done.