Time Travel: 2014

Chapter 232 Tens of Millions of Taxes \u0026 Social Responsibility

The revenue of the game "Clash of Clans: Clash Royale" exceeds 3 billion US dollars.

Under such circumstances, if Lin Hui can really create such a super cash cow, it seems unreasonable not to divide it into a billion dollars.

Of course, you need to pay attention to the strategy in actual implementation. Tell the super cells directly that Lin Hui will get more than one billion US dollars. They will definitely not foolishly agree to give up one billion US dollars.

Even if the managers of Super Cell are so stupid, the scumbags behind Soft/Bank probably won’t either.

Then some planning will be needed.

But it’s useless to think about it for the time being.

Many things were not entirely based on Lin Hui's will.

…

With nothing to do in the morning, Lin Hui went to find Eve Carly again.

As he was about to go abroad, Lin Hui felt that getting in touch with some Wai Guo Keren would help improve his speaking skills.

In addition, yesterday Lin Hui and Eve Carly only talked about a basic framework.

There are still many things that need to be discussed in more depth.

It happened that Eve Carly was also willing to communicate with Lin Hui, so the two met again at the Number One Scholar Academy.

But this time Huang Jing and Mina Kali are not here.

Even though a man and a woman were alone in the same room, no one else would have come up with any strange ideas while talking about processing high-dimensional data with a member of the opposite sex all morning.

When communicating with Eve Kali this time, Eve Kali still talked the most.

Lin Hui plays more of a listener role, judging from Eve Carly's introduction.

Eve Carly understood quickly. She had already roughly understood what Lin Hui had told him before about applying distributed word vectors to process raw data.

And initially based on the theory introduced by Lin Hui before, we made a model of distributed word vectors.

Although this model is not very efficient yet, it has already taken shape.

Eve Carly also realized that this model was not very efficient. She didn't understand how to achieve the efficiency as described by Lin Hui.

Lin Hui saw the model created by Eve Carly and immediately understood why Eve Carly's model was not efficient enough.

Because Eve Carly developed a static distributed word vector model, the earliest distributed word vector model born that year was a static model.

But the static model is not a dynamic distributed word vector model after all.

The efficiency of the static model is somewhat worse than the dynamic distributed word vector model.

But it’s not her fault that Eve Kali failed to think of building a dynamic distributed word vector model.

To a large extent, this is due to the limitations of people's thinking in a certain time and space or a specific era.

Lin Hui could give some advice to Eve Carly.

But Eve Carly stubbornly said she had to think about it first.

She is really a stubborn woman, but Lin Hui respects her ideas.

Eve Carly was curious about how the language model of the generative text summarization algorithm Lin Hui had previously developed was trained.

This involves the distinction between supervised learning and unsupervised learning.

Generally speaking, methods based on supervised learning involving natural language processing first need to use artificial or semi-artificial methods to construct a corpus of documents in the field of study, and then select appropriate features to train the corresponding classifier or sequence annotator. Then use the trained classifier or sequence annotator to label each sentence of the new document, and extract the sentences marked as candidates and splice them into a summary.

Methods based on unsupervised learning do exactly the opposite. Unsupervised learning does not require artificial or semi-artificially labeled corpora.

In this way, the workload based on unsupervised learning is often much smaller than that based on supervised learning.

The above is just a little difference between supervised learning and unsupervised learning. The deeper difference between the two cannot be explained in one sentence or two.

As it turned out, Lin Hui was able to escape after chatting with Eve Carly until about three o'clock in the afternoon.

Regarding the takeaway from this conversation, it seems that Eve Carly started calling Lin Hui "teacher."

Lin Hui: (⊙﹏⊙), the number of strange identities has increased.

…

In the afternoon, Lin Hui received another text message from the bank.

Judging from these notifications, the balance on Lin Hui’s 6699 card changed from 57.5 million yuan to 73.5 million yuan.

The reason why the domestic balance has grown so rapidly may be related to the situation Huang Jing previously reported to Lin Hui.

Many overseas players use Chinese accounts to download games due to differences in pricing.

Sure enough, Lin Hui noticed that the balance on 0010's card had almost no change, but increased from US$36.77 million to US$36.9 million.

This growth rate is not even comparable to Lin Hui’s corresponding card with overseas online articles.

The card corresponding to the overseas web article also increased by US$190,000, and the balance changed from US$660,000 to US$850,000.

Of course, this is also because South Wind Pro has not yet reached the 24-hour settlement cycle, otherwise the overseas application share would not be only more than 100,000 US dollars.

The changes in these data were basically within Lin Hui's expectation.

Although these incomes were generally within Lin Hui's expectations.

But Lin Hui still logged into the app store and checked the corresponding data details from the developer backend.

The reason why we need to look at the corresponding data is not because Lin Hui has to carefully calculate his income.

It's because a lot of information can be deciphered through the background data.

In this sense, data is a silent but direct feedback to developers.

Timely analysis of background data and making corresponding adjustments to the online software based on the information interpreted from the analyzed data can enable developers to better cater to the market and meet user needs.

Judging from the background data, Lin Hui’s total pre-tax income in the past 24 hours was 19,047,620 yuan (after-tax income was 16,000,000.80 yuan)

He paid more than three million yuan in taxes. Lin Hui roughly estimated that from June 17 to now, he had paid tens of millions in taxes in less than ten days.

And these are only the taxes involved in domestic income.

As for foreign income, if it is now settled together, the amount of tax paid will be at least several times that of domestic income.

Lin Hui remembered that he spent US$1 million to promote the game "FLAPPYBIRD" overseas. He spent US$65,000 to build the greyforest.net website, and spent a total of US$8 million to buy domain names, which involved overseas patents for generative summary algorithms. The application cost approximately US$200,000, and the patent acquisition cost US$1.65 million.

At this time, the balance on the card corresponding to Lin Hui Software’s income was US$36.9 million.

A rough calculation shows that Lin Hui earned a total of US$47.815 million overseas through the App Store.

According to the exchange rate of 6.15, these overseas income correspond to 294,062,250 yuan.

If the individual tax declaration is based on royalties, Lin Hui will have to pay a tax of 47.0499 million yuan.