Chapter 257: I’ve Seen Plenty

Chapter 257 -194: I’ve Seen Plenty

Chen Lin saw the excited gleam in Pan Si’s eyes. He organized his thoughts and slowly began to speak:

“Dr. Pan, I have to remind you of something.”

“The algorithm I derived at the mathematics conference was publicly released.”

Chen Lin’s tone was calm, but his words were clear.

“Furthermore, I’m going to compile it into a formal paper and publish it in JAMS.”

“I don’t plan to use patents or any other means to restrict the use of this algorithm.”

Hearing this, Pan Si was slightly taken aback.

Chen Lin continued:

“Based on what I’ve seen online these past few days, AI teams all over the world—whether it’s giants like CloseAI and Doggie, or labs at various universities—are all trying to integrate this algorithm into their existing model training pipelines.”

“So,” Chen Lin said with a shrug, “when it comes to applying this algorithm, our new team probably won’t be the first.”

After Pan Si heard this, the excitement on his face faded a little.

He pushed up his glasses and nodded thoughtfully.

“I understand, President Chen.”

“It’s true. If the algorithm is open-source, a startup team like ours definitely can’t compete with the big companies in terms of application speed.”

Seeing the disappointed look on Pan Si’s face, Chen Lin immediately changed his tune:

“But, Dr. Pan, you don’t need to worry.”

His voice was filled with confidence.

“I’ll be starting new mathematical research right away, trying to develop newer and better artificial intelligence algorithms.”

“When the time comes, we can use the latest breakthroughs to train our own model.”

At these words, Pan Si’s eyes lit up again.

“President Chen, you mean… you’re going to continue researching AI algorithms?”

“That’s right.” Chen Lin nodded. “Mathematical research has always been my strong suit.”

“Since we’re going to build a large AI model, I will definitely provide as much support as I can on the algorithmic level.”

After hearing this, Pan Si became completely energized.

“That’s fantastic!”

He slapped his thigh forcefully.

“President Chen, if you can continuously produce new algorithms, then our project will have a huge advantage!”

“You have to understand, what the industry lacks most right now are talented people who can innovate on a mathematical level.”

“Most AI companies are only making optimizations in engineering and implementation. There’s almost no one who can truly break through the attention mechanism on a theoretical level.”

As Pan Si spoke, his tone became extremely serious:

“If you can do that, President Chen, then I’m in!”

A smile tugged at the corner of Chen Lin’s mouth when he heard this.

“Then as for compensation, I assume you’re clear on that, Dr. Pan?”

“I am.” Pan Si nodded. “A hundred thousand a month, after taxes. I can accept that.”

“However,” he paused, then added, “I hope to get some company stock options.”

“After all, if this project really succeeds, the company’s valuation will definitely skyrocket.”

“I want a piece of the pie, too.”

Chen Lin wasn’t surprised by this request.

Someone like Pan Si, a returnee who had worked at a major foreign company, would naturally be well-versed in how equity incentives work.

“Done.” Chen Lin agreed readily. “We can discuss the specific percentage of options in detail later.”

“But there’s one thing I have to make clear upfront—the options will have a vesting period and will be tied to the project’s milestones.”

Chen Lin had already discussed this point with Huang Yiyi. It was acceptable to offer equity incentives to top technical experts willing to join Yanhai, as long as Chen Lin could buy back the shares at market price after they were exercised.

For Chen Lin, equity was very important. After all, it could be directly counted as a personal asset by the [Little Financial Planner].

Upon hearing this, Pan Si immediately broke into a smile.

“Of course. I understand.”

The two of them discussed a few more details and finally finalized the arrangements for him to join the company.

Pan Si said that the soonest he could officially start would be next Monday.

After seeing Pan Si out, Chen Lin returned to his desk and let out a long sigh of relief.

‘Bai Qian and Pan Si, two top tech experts, are both on board!’

‘Next, it’s time to officially kick off the AI project.’

…

However, Chen Lin soon discovered that things wouldn’t go as smoothly as he had imagined.

「Monday morning, November 10th.」

In the Yanhai Company conference room, Chen Lin, Bai Qian, and Pan Si sat together for the first time.

The conference room wasn’t large. There was a long rectangular table surrounded by a dozen or so chairs.

At that moment, Chen Lin sat at the head of the table, with Bai Qian and Pan Si on either side of him.

Sun Yu, Huang Yiyi, Yunqing, and Shen Yan sat a little farther away, observing the meeting.

Chen Lin glanced around the room and was the first to speak:

“I’ve called everyone here today mainly to discuss the technical roadmap for our upcoming AI project.”

“Dr. Bai, Dr. Pan, you are both experts in this field, so I’d like to hear your opinions.”

Bai Qian and Pan Si glanced at each other, then spoke at almost the exact same time:

“I’ll go first—”

After speaking, they both stopped at the same time.

The atmosphere instantly became a bit awkward.

Seeing this, Chen Lin smiled and smoothed things over:

“Why don’t you go first, Dr. Bai?”

Bai Qian nodded, pushed up his glasses, and cleared his throat.

“Then I’ll be direct.”

He took a tablet out of his backpack, opened a PPT file, and projected it onto the large screen in the conference room.

“I believe that if we’re going to build a brand-new large AI model, the first thing we must do is innovate on the architecture.”

A complex architectural diagram appeared on the screen.

Bai Qian pointed to several modules on the diagram and began to explain in detail:

“The current mainstream Transformer architecture, despite its advantages in parallel computation and capturing long-range dependencies, has very low parameter efficiency.”

“Moreover, the Transformer’s self-attention mechanism has a computational complexity of O(n2), which causes the computational cost to grow exponentially when processing long texts.”

“Therefore,” Bai Qian’s tone became incredibly firm, “we must design a brand-new architecture to solve these problems.”

He clicked to the next slide of the PPT, which was filled with a slew of mathematical formulas and architectural diagrams.

“What I’ve been researching for the past few years is a new type of neural network architecture based on a dynamic sparse activation mechanism.”

“To put it simply, it allows the model, during inference, to not activate all of its parameters. Instead, it dynamically selects which parts to activate based on the input.”

“This way, computational efficiency can be improved severalfold, and the model’s capacity can also be greatly increased.”

Bai Qian spoke with great passion, radiating the aura of a true scholar.

However, the expression on the face of Pan Si, who was sitting opposite him, grew increasingly strange.

As soon as Bai Qian finally finished speaking, Pan Si immediately spoke up:

“Dr. Bai, with all due respect.”

His tone was polite, but the meaning behind his words was anything but.

“I’ve seen too much of this kind of stuff in academia.”

“Every year, countless papers claim to have designed some ‘revolutionary’ new architecture.”

“But what happens in the end?”

Pan Si spread his hands, a mocking smile on his face.

“The vast majority of them only ever run on toy models in a lab. They can’t be applied to actual large-scale model training.”

“Why?”

He didn’t wait for Bai Qian to answer, continuing on his own:

“Because these so-called ‘new architectures’ often look great in theory, but in terms of engineering and implementation, they’re either unstable during training, too slow during inference, or have excessively high hardware requirements.”

“In short, they’re just not practical!”