Returning to his senior year of high school, he went all in and made billions by buying at the botto

Chen Mo stayed at Tsinghua University until the end of the month, while in Suzhou and Hangzhou, Bian Xuezhang and Liang Wenfeng reported that the model had successfully run.

Chen Mo immediately took a private jet to Suzhou-Hangzhou Airport.

As I stepped out of the business jet, a light rain had just fallen.

The vehicle for Deep Dive is already waiting outside the tarmac.

Bian Xuezhang sat in the back row, wearing a plaid shirt.

He looked much more energetic than he had a few months ago.

"Mr. Chen," he said, opening the car door, "the model has run successfully."

"Go to the company first."

Chen Mo got into the car.

Bian Xuezhang immediately turned on his portable computer.

"The first-generation model completed its final round of live trading tests three days ago."

……

Deep Exploration is located in the Suzhou-Hangzhou Riverside R&D Park.

The initial funding of 200 million yuan allowed Deep Exploration to skip the most difficult initial stage.

Liang Wenfeng did not use the money to decorate his office.

Apart from a simple rest area, almost all the investment went into servers, data, talent, and trading systems.

In the conference room.

Liang Wenfeng and Xu Jin had been waiting for more than half an hour.

There were three reports on the table.

Model backtesting.

Out-of-sample validation.

Small-scale live trading.

After Chen Mo sat down, Liang Wenfeng pushed the top dish directly in front of him.

"A generation model completed by Bian Xuezhang and his team."

"Internally, we'll temporarily call it DS-Q1."

"It mainly consists of four parts: multi-factor stock selection, market-neutral hedging, volatility control, and trade execution."

Historical backtesting began in 2007.

"No future data was used; trading suspensions, price limits, trading impact, and transaction fees were all calculated using conservative estimates."

Chen Mo opened the report.

In the past eight years of historical backtesting, the model has achieved an annualized return of over 30%.

The maximum drawdown should be kept below 9%.

What's really important are the out-of-sample results from the last few months.

Once the main framework of the model is determined, the test data will no longer be used for parameter training.

Even after experiencing continuous market declines, liquidity contraction, and several sharp fluctuations, the strategy still maintained positive returns.

"A 100 million yuan live trading test account will be operational starting in August."

Liang Wenfeng said.

"As of this week, the cumulative return is 12.8%."

"Maximum drawdown of 2.7%"

"The stock portfolio itself was affected by the market downturn, but index hedging and position control offset most of the systemic risk."

"Currently, the correlation between the model and the overall market has been reduced to a very low level."

Chen Mo continued flipping through the pages.

During the most extreme days, the first-generation model also experienced losses.

It cannot accurately predict every drop, nor can it avoid all false signals.

However, when the market environment changes, the model will proactively shrink its exposure and reduce the weight of failure factors.

Being able to make money is very important.

Knowing when to bet less is equally important.

"What about the capacity?"

"The preliminary estimate is between 1.2 billion and 1.5 billion yuan."

Bian Xuezhang answered.

"Once the figure exceeds 1.5 billion yuan, the impact cost of some small and medium-sized stocks will rise rapidly."

"If we only use stocks with better liquidity and reduce the turnover rate, the capacity can be expanded to about two billion yuan."

"But profits will also decrease."

One billion is the most comfortable scale at present.

It can verify the performance of the strategy under large funds without immediately hitting the capacity limit.

"What about risk control testing?"

"I've worked through six different extreme scenarios."

Liang Wenfeng took over the conversation.

"A sudden drop in stock index futures liquidity, widespread suspension of trading in individual stocks, distortion of hedging basis, interruption of market data, delays in trading channels, and model failure for five consecutive trading days."

"If any of these factors reach a preset threshold, the system will automatically reduce the position size."

"If multiple risks occur simultaneously, the trading authority will be transferred to manual risk control."

Chen Mo closed the report.

"We can ship the products now."

The meeting room was silent for a few seconds.

Although Liang Wenfeng was prepared, he still couldn't help but feel his breathing become heavy.

Since its inception, Deep Search has been using its own funds and internal testing accounts.

The issuance of the first formal fund means that the market will begin to test this model with real net asset value.

What is the target size?

"One billion yuan."

Chen Mo answered.

Liang Wenfeng looked at him.

"We originally planned to raise only 500 million yuan."

"The capacity of the first-generation model is sufficient."

"One billion yuan was subscribed by Taichu."

"Tai Chu No. 3 and No. 4 will each contribute 500 million yuan."

The three of them stared at Chen Mo in shock.

The first fund.

The entire 1 billion yuan was subscribed by Taichu.

Deep exploration doesn't even require finding a single client.

"A single injection of one billion yuan will impact the strategy."

However, Liang Wenfeng was not blinded by his emotions.

"A 100 million yuan test account and a 1 billion yuan official product are not the same thing."

"The subscription amount is confirmed at once, and the funds are entered in stages."

Chen Mo said.

"The first phase will not exceed 500 million yuan."

"After running for a month, gradually increase the amount based on the impact of trading volume, factor decay, and actual trading pullback."

"If the model does not perform as expected, the remaining funds will remain in the escrow account. We cannot forcibly expand the capacity just to use up all the money."

Liang Wenfeng nodded.

"That's fine."

There is a connection between the beginning and the deepest search.

Chen Mo is both the actual decision-maker of Taichu and holds 80% of the shares of Deep Search.

It's impossible for a 1 billion yuan subscription to be transferred directly based on just one sentence from him.

Before setting off for Suzhou and Hangzhou, Taichu had already initiated a review of related-party transactions.

Chen Mo abstained from voting.

Zhao Qiang, Xu Xiang, independent risk control personnel, and the custodian jointly reviewed the strategy report, live trading records, valuation plan, and fee structure.

Taichu No. 3 allocated 500 million yuan from its long-term strategic allocation.

Taichu No. 4 will allocate 500 million yuan from its absolute return strategy quota.

The combined investment in the two products accounts for only five percent of their respective scales.

Deep Search also waived the right to charge repeated fixed management fees for the 1 billion yuan, and only extracted performance-based compensation according to the agreement.

All related parties, fees, and risks will be fully disclosed to the investors of both funds.

Xu Jin placed another document on the table.

"The official name is Deep Exploration Quantitative No. 1 Private Equity Investment Fund."

"The target size is one billion yuan."

"If Taichu confirms full subscription, we will no longer accept other funds."

"Full capacity."

Chen Mo picked up a pen and signed his name on the subscription opinion form.

"Let's look at the net asset value for the first year first."

"The market will likely be scrambling to take the remaining shares."

……

With the funding issue resolved, Bian Xuezhang did not leave the meeting room immediately.

He turned the computer back on.

"Mr. Chen, there's one more thing."

A new set of model structures appears on the screen.

Unlike the first-generation quantitative model, this time there is no clearly arranged factor table.

A large amount of market data, financial data, cash flow data, and news data are fed into a more complex learning system.

"Most of the features of the first-generation model are still defined in advance by researchers."

Bian Xuezhang said.

"We tell the program what valuation, momentum, volatility, and trading structure are, and then let it find the most effective combination."

"The problem is that the factors that humans can think of are always limited."

"Some variables may seem meaningless in isolation, but when placed in a specific market environment, they can lead to completely different results."

Liang Wenfeng took over the conversation.

"We want the model to find relationships in the data itself."

"It no longer just answers whether a factor is effective, but also identifies what state the market is currently in."

Chen Mo looked at the screen.

This is precisely the path that quantitative investment will inevitably take in the future.

Machine learning, deep learning, natural language processing.

Market trends are only one part of the market.

News, announcements, changes in the supply chain, and subtle attitudes expressed by company management can all be transformed into information.

As computing power and modeling capabilities continue to improve, quantization systems will no longer only process numbers.

"You want to do AI quantitative analysis."

"right."

Liang Wenfeng's eyes lit up.

"It's just a concept right now."

"There are problems with computing power, training methods, and data labeling."

"If we directly let complex models predict stock prices, we are likely to get an overfitting result that is even harder to detect."

"Therefore, the first step is not to predict whether prices will rise or fall."

Bian Xuezhang said.

"First, let it perform market condition identification, factor screening, and trade execution optimization."

"The model provides suggestions, but the final risk exposure is still controlled by the existing system."

Chen Mo nodded.

The two were not prepared to hand over all their funds to an inexplicable black box simply because of the name "artificial intelligence".

They hope to use AI to address the aspects that humans are least good at first.

Finding complex relationships from massive amounts of data.

Then keep the boundaries of funding and the final decision-making process within a controllable range.

"The direction is correct."

Chen Mo said.

"A separate AI quantitative research group can be established."

"The server, GPU, and data budgets have been resubmitted and will not be included in the day-to-day R&D budget for the first-generation model."

Liang Wenfeng looked at him: "Mr. Chen is very knowledgeable about artificial intelligence?"

"I helped found an artificial intelligence research institution in the United States."

"OpenAI".

The people in the meeting room fell silent at the same time.

Since its inception, OpenAI has attracted considerable attention in the global artificial intelligence research community.

Open research, long-term intelligence, and shared computing power.

The addition of Musk and many other Silicon Valley tech figures has placed it in the spotlight since its inception.

Deep Exploration has also been paying close attention to this organization.

Liang Wenfeng had no idea that Chen Mo was one of the initiators behind it all.

Did you found OpenAI?

"I'm not the only one who initiated this."

Chen Mo said.

"I will provide the main funding, computing power, and some research directions."

Liang Wenfeng's interest grew: "What are they mainly researching now?"

"General training frameworks, reinforcement learning, language intelligence, and how to enable models to complete complex tasks with fewer manual rules."

Chen Mo did not elaborate on the specifics of the internal projects.

"Some of these general methods are similar to the problems that quantization models need to solve."

"DeepExploration can join OpenAI's collaborative research network to share publicly available research results, some training frameworks, and computing tools."

"You can also send someone to participate in the internal seminar."

Bian Xuezhang asked, "Does the transaction data need to be uploaded?"

"unnecessary."

Chen Mo's answer was very clear.

"The account data, strategy parameters, and real-time trading records of Taichu and Deep Quest do not leave the country."

"OpenAI shares general artificial intelligence methods."

"Deep exploration can provide the organization with basic research results that do not involve trade secrets."

"The boundaries of code, data, and intellectual property rights between the two parties are written into the agreement in advance."

Liang Wenfeng did not speak immediately.

He looked again at the words "Deep Exploration".

When Chen Mo first wrote down this name, he felt it was more like a laboratory.

A few months later, the company had just completed its first-generation quantization model and obtained a ticket to join the global artificial intelligence research network.

"I'll participate."

Liang Wenfeng raised his head.

"I will personally lead the AI ​​Quantitative Research Group."

Bian Xuezhang pushed up his glasses.

"I'm responsible for organizing the parts that can be disassembled from the first-generation model."

Xu Jin also added a budget line to the paper.

The team quickly broke down the task into five parts: data, computing power, models, risk control, and personnel.

Chen Mo sat to the side, listening to the three of them divide the tasks.

In its previous life, the pursuit of depth initially relied on quantitative trading to accumulate funds, talent, and computing power.

Many years later, we finally moved towards a large-scale model of artificial intelligence.

With Chen Mo's help in this life, funding, technology, and access are not problems.

What would have taken many years can be accelerated starting today.