Chapter 256: Complementary

Chapter 256 -193: Complementary

“It’s alright.” Chen Lin smiled and waved a hand. “I can tell you’re very passionate about this field.”

“That’s a good thing.”

He paused for a moment, then continued:

“Dr. Bai, here’s what I’m thinking.”

“If you’re willing to come work for Yanhai, I’ll provide you with ample resources and the freedom to bring your ideas to life.”

“Computing power, data, staffing—none of that will be an issue.”

“My only requirement is that you do everything you can to help me make this project a success.”

Bai Qian was completely stunned when he heard this.

His eyes widened as he stared at Chen Lin in disbelief.

“President Chen, are… are you serious?”

“Of course.” Chen Lin nodded. “I never joke.”

Bai Qian swallowed hard, his voice trembling slightly:

“And as for the compensation…”

“Eighty thousand a month, after taxes,” Chen Lin stated plainly. “And if the project is a success, there will be an additional project bonus.”

Bai Qian was completely dumbstruck when he heard the figure.

Eighty thousand a month, after taxes!

That was so much higher than his income at the Institute of Automation!

He took a deep breath, trying to calm himself down.

“President Chen, could you give me some time to consider?”

Chen Lin smiled. “Of course. A decision this important certainly requires time to think over.”

“However,” he paused, “I do hope you can give me an answer soon. The project is already in the preparatory stages, and the sooner we launch, the better.”

Bai Qian nodded vigorously. “I understand! I’ll give you my answer as soon as I can!”

After a few more words, Bai Qian stood up to take his leave.

Chen Lin walked him to the door. As he watched Bai Qian’s retreating figure, a satisfied smile touched his lips.

‘This Bai Qian… he’s a true tech heavyweight.’

‘If I can get him on board, it’ll be a massive boost for the project.’

…

…

「The next day, Tuesday afternoon.」

Another guest arrived at Yanhai Company’s reception area.

Chen Lin was still seated on the sofa. Opposite him was a man who looked to be in his early thirties, wearing gold-rimmed glasses and a well-tailored casual suit.

This was Pan Si, the high-flying expert recommended by President Peng, who had recently returned from working abroad.

A stark contrast to Bai Qian’s typical tech-nerd disposition, Pan Si exuded the mature, confident, and effortless aura of a seasoned corporate professional.

Upon meeting, Pan Si took the initiative, extending a hand with a perfectly composed smile:

“President Chen, I’ve heard so much about you! A pleasure to finally meet you in person.”

Chen Lin smiled as he shook his hand. “Dr. Pan, you’re too kind. The honor is mine. It’s not every day you meet an AI expert who’s worked at Doggie.”

Pan Si smiled and waved a dismissive hand. “You’re too generous. I was just a rank-and-file employee at Doggie, nothing to write home about.”

After the two were seated, Yunqing came over with tea.

This time, however, when Pan Si took the teacup, he acted very naturally and even politely said “thank you” to Yunqing.

Yunqing smiled, nodded, and turned to leave.

Chen Lin picked up his teacup, took a sip, and was the first to speak:

“Dr. Pan, let’s not beat around the bush. Let’s get straight to business.”

“I assume President Peng has already given you the general gist of the situation here, correct?”

Pan Si nodded. “Mm, President Peng filled me in. You intend to develop a brand new large AI model in-house, correct?”

“That’s right.” Chen Lin nodded. “And I want this model to surpass the current mainstream large models on the market in every aspect.”

Upon hearing this, Pan Si smiled, a hint of teasing in his tone:

“President Chen, that’s no small goal you’ve set.”

“You have to understand, there are hundreds, if not thousands, of companies worldwide working on large AI models right now.”

“To ‘surpass them in every aspect’ is no small feat.”

Chen Lin also smiled. “I know it’s difficult, but I believe that as long as the direction is right, and the funding and talent are in place, it can definitely be done.”

Pan Si gave a slight nod, offering no rebuttal.

He pondered for a moment before speaking slowly:

“President Chen, may I ask if you have any specific ideas about the model’s technical roadmap?”

Chen Lin shook his head candidly. “To be honest, I’m not an expert in the field of AI. That’s why I’d like to hear your thoughts.”

Pan Si’s eyes lit up slightly when he heard this.

“Then I’ll be direct.”

He pushed up his glasses, his tone growing serious.

“In my years working at Doggie, I’ve participated in the training projects for several large models.”

“Based on my experience, I believe that the current mainstream model architectures in the industry, like Transformer, MoE, and so on, are actually sufficient.”

“They have all been proven effective in theory.”

“What truly determines a model’s performance isn’t the architecture itself, but two other things—”

Pan Si held up two fingers.

“First, data quality.”

“Second, alignment.”

Chen Lin raised an eyebrow slightly upon hearing this.

“Can you elaborate?”

“Of course.” Pan Si nodded and went on to explain:

“Data quality refers to just how clean and high-quality the data you use to train the model is.”

“Many people think that building a large AI model is just about stuffing massive amounts of data into it and letting the model learn on its own.”

“But in reality, if your data contains a lot of noise, errors, and biases, the resulting trained model will definitely not perform well.”

“So, data cleaning is a critically important step.”

“In fact,” Pan Si’s tone became incredibly serious, “you could say that data quality determines more than 50% of a model’s success.”

Hearing this, Chen Lin nodded thoughtfully.

Pan Si continued:

“As for alignment, that refers to making the model’s output conform to human values and expectations.”

“For example, when you ask the model a question, the answer it gives you should be useful, accurate, and harmless.”

“Rather than spouting nonsense or outputting harmful information.”

“This requires using techniques like RLHF (Reinforcement Learning from Human Feedback) to conduct alignment training on the model.”

Pan Si reached this point, then his tone suddenly shifted.

“However, President Chen, the algorithm you derived at the mathematics conference has actually solved the problem of model hallucination on a mathematical level.”

“This is equivalent to improving the model’s reliability directly at the source.”

“So, if you could integrate that algorithm into the model training…”

A flash of excitement shone in Pan Si’s eyes.

“I believe that the model we train will definitely be able to surpass all competing products on the market in terms of content generation quality and reliability!”

As Chen Lin finished listening to Pan Si, something clicked in his mind.

‘He suddenly realized that although Bai Qian and Pan Si had different views on model architecture, their perspectives were actually complementary.’

‘Bai Qian is more focused on innovating the model architecture, while Pan Si is more focused on data and alignment.’

‘If I could get them both on board, and let each of them play to their strengths…’