Original source: Beluga whale out to sea
![]()
Image source: Generated by Unbounded AI
After a weekend of viral rumors, on Monday, the news that “the vast majority of employees will be laid off” is basically an official announcement. “There will indeed be business direction and organizational adjustments, and more emphasis will be placed on the exploration of some innovative games and related technologies”, this is the official reply given by Chaoxi Lightyear.
Byte’s abandonment of the game business can be described as a resolute one.
There are also various conjectures about the reason. Some people think that games can’t be sustained as a cash flow business line, and at the same time, there is no room for imagination, and Byte is going back to do AI;Some people think that Tencent and Byte shake hands and make peace, after Byte gives up the game business, Tencent games will no longer have a psychological burden to buy volume from bytes for live broadcasts;Even,There are even those who say that Zhang Yiming doesn’t like games…
No matter how you say it, we think it still makes some sense, anyway, the last one is definitely not Zhang Yiming’s standard for judging business logic.
Interestingly, on November 27, 2023, the same day that it was confirmed that Byte would abandon the game business, 36kr reported that ByteDance recently established a new AI department, Flow, with the technical leader being Hong Dingkun, vice president of technology at ByteDance, and Zhu Wenjia, the head of the business team. In the recruitment post released by Flow, Flow said that it is the AI innovation business team of ByteDance, and has launched two products, Doubao and Cici, respectively in China and overseas, and a number of AI-related innovative products are being incubated.
On the same day, Byte completed the -1 & +1 of the department on various network information channels.
Cici is on the overseas climbing list, but a GPT casing application?
In fact, judging from the situation of Doubao and Cici, it should have been a while since Byte established Flow, or the business has been going relatively smoothly, and now a department has been officially established.
In February 2023, Byte began to set up a large model team to work on two modalities: language model and image, with the language model team led by the search department and the image model led by the intelligent creation team.
At the beginning of June, Byte was in the closed beta of the AI dialogue project codenamed Grace, and on August 17, Grace changed its name to Doubao. Cici was also launched in a number of overseas markets in August. In view of the fact that the environment is not friendly to AI bots in China, and Byte is also a strong global gene, overseas should be a more powerful market for Byte.
Diandian data shows that Cici has been launched in 36 overseas countries and regions, basically except for mature markets such as the United States and Europe, almost all regional markets are covered, including Japan, South Korea, the Middle East, Southeast Asia, Africa, and South America, such as Saudi Arabia, Indonesia, Brazil, etc.
![]()
From the perspective of mainly doing emerging markets, apps are also mainly focused on the Android side, whether it is the time to start the launch, the number of materials to be delivered, or the continuation of the delivery efforts, the Android side is more valued.
![]()
In the past few days, Cici has begun to climb the list, and on the 28th, he won the first place in the overall list of Google Play Argentina and Peru, and in the Philippines and Indonesia, he also hovered around the top 10 of the overall list.
In terms of positioning, Cici focuses on “Your Helpful Friend” + “Free”. And Bytes, which has its own large model lark, appeared in Cici’s Southeast Asian materials with the copy Powered by GPT-4. So, is Byte doing a charity business overseas to allow users in emerging markets to use GPT-4’s services for free?
![]()
To verify this, we downloaded Cici to test a wave.
GPT3 + search, product pass overall
Let’s answer the above question first.
When I asked Cici, “What large model did you use to train it?”, Cici’s answer instantly slapped the creative in the face, “I trained it based on OpenAI’s GPT-3 model.” And I asked, “Are you sure?” and Cici replied in the affirmative again.
![]()
However, when I asked, “Do you know about ChatGPT-4” and “What happened in November 2023?”, Cici can also answer.
If Cici does not have “hallucinations”, then the developer should be based on GPT-3 for training + search engine. Based on the few questions I’ve asked, when Cici gives a search answer, it seems that it also provides users with a source of information.
Despite using an older version of the model, Cici’s strength is okay.
Cici is available in 12 languages, barely covering the Middle East, Southeast Asia, and Japan and South Korea. ChatGPT is currently mainly available in the languages of some European countries, Japanese, Korean, and Russian, and Cici has done corresponding localization work for the app’s market planning. The user can set the language in the upper right corner of the dialog with Cici, but if the user switches the language of the question directly in the dialog, Cici can also switch the language directly.
![]()
After the user enters the dialog box, Cici enables voice reading by default, and after the user sends a question, Cici will read the text again with voice at the same time.
In addition, the app will also provide some chatbots tailored to the needs of segments, such as tarot measurement, language learning, copywriting, programming, and so on. Users can also create their own chatbots, which they choose to use publicly or only for personal use.
![]()
In order to determine whether Cici is acting or not, I tested it from two dimensions, one is the basic functions, including the speed of response, the ability to understand multiple rounds of conversations, and the ability to identify sensitive content. The other is the amount of knowledge reserve, to see if it can solve various problems and the depth of the answers.
**1. Response time: Fast. **
If the answer is answered by a large model, Cici is almost a second. If a search is required, Cici can also respond within seconds.
**2. Multi-round conversation comprehension: Depends on the number of conversational rounds. **
If a few questions are not too far apart, Cici can remember the context. For example, when my first question is what is the Gini coefficient for Japan, and the second question is immediately after “What about Korea?”, Cici can understand that I am asking about the Gini coefficient of Korea.
![]()
And when I asked Cici earlier to call me Cecilia, Cici said yes, but after a few rounds of irrelevant questions, when I asked “what’s my name”, Cici still had a hard time answering.
![]()
ChatGPT is significantly better at this, and when I ask it to answer a question from a few days ago in the same dialog, it only needs to prompt the keyword, and ChatGPT can understand which question I’m talking about.
![]()
**3. Sensitive Content Recognition Ability: Pass. **
![]()
When I asked Cici some sensitive questions, Cici was able to identify sensitive words and refused to answer the questions head-on. Below the answers, Cici will also set up some relevant questions to try to guide users to think positively. It’s just that compared with ChatGPT3.5, Cici’s replies are more routine, while ChatGPT3.5 is more human.
![]()
In fact, from the several default segmented chatbots provided by the developer, it can be seen that Cici has dabbled in all kinds of knowledge, and is not as good at liberal arts content as Baidu Wenxin Yiyan was when it was first launched. In terms of the depth of the questions, Cici answered more specifically on some topics.
Taking language learning as an example, I asked Cici to teach me Chinese in the English context, and Cici could accurately point out grammatical errors, and although in the English system language, the voice function would automatically skip Chinese, but when I asked Cici to teach me how to pronounce it, it would mark pinyin in parentheses. ChatGPT’s answer to the same question is much more sloppy.
![]()
![]()
Cici’s answer (top) ChatGPT (bottom)
Cici, this overseas chatbot app, although it is not a simple call to the API, but the product is still very rudimentary. At this point in time, what is the thinking behind Byte’s promotion of such a decent AI conversation app to climb the list in overseas markets?
AI path conjecture for bytes
1, AI bot is the traffic entrance, return to the industry
Byte’s trump card products are all entertainment or content products, but Byte itself is a platform, and does not produce content, mastering the traffic entrance is what Byte is better at, and AI bot is at least the most clear new traffic entrance in recent years.
Looking at ChatGPT alone, before Meta launched Threads, ChatGPT broke TikTok’s record, and the number of users exceeded 100 million in just 2 months after its release.
![]()
Image source: Yahoo Finance
Since then, there have been countless apps in the app store that call the GPT interface to make chatbots directly, as well as apps that integrate chatbots into existing products.
2. Perfecting the technology in the toC again and again
Now the consensus is that there is still a long way to go between domestic large models and GPT. Crashing on a large model sounds like an option with a very low ROI. The establishment of the Flow department is also a force application layer.
At the same time, bytes will also have a success path dependency. The recommendation algorithm has always been something that Byte is proud of, which makes Meta very greedy, and we do not rule out that the Byte algorithm started very high, but the follow-up relies on a large number of user behavior and data feedback to continue to improve.
The same applies to AI models, if it is really a GPT-3 model as Cici said, it is a clear idea for Byte to enter the market first to reach out to users.
3. Positioning differences
At present, making free AI bots in emerging markets is to avoid the edge of ChatGPT (but based on the premise that GPT-4 is free to use), and it is also different from many small and medium-sized companies that do AI bots, they mainly rely on subscription monetization, which is a typical App designed through apps and earning the difference, of course, Cici currently seems to be too.
If Byte’s miracle trick is repeated again, the promotional version allows users in emerging markets to use GPT-4 for free, which can avoid the products on the market and quickly acquire a large number of users.
In fact, the current call cost of GPT-4 API is still quite high, especially the difference between the cost of advanced large models and some open source large models can be hundreds of times, after Turbo came out, the price was lowered a little, but many developers said that the call cost is still very expensive, and it is also higher than the previous 3.5.
If Byte can adjust the model and get a better experience, on the basis of acquiring a large number of users, advertising and monetization are not unacceptable.
![]()
After today’s news came out, some netizens ridiculed “It’s only now that AI is deployed, and then it will be withdrawn when the products of friends are launched?”, but I have to say that even so, we may also tend to think that bytes do AI applications, at least compared to games, which sounds more reliable…
