Chapter 41: The Big and Small of Future Models
"Meng didn't consider joining whiteness at all?" Robin Li's expression fluctuated for a moment, but it was almost undetectable, "I think I should be the CEO who attaches the most importance to AI technology in the whole of China now, and our whiteness is also the company with the largest investment in the whole of China." ”
"You've got a place to go? Or do you have high requirements for the conditions? We can talk about the specific situation again, whether it is too hasty to skip this link directly. Li Yanhong's heart to seek talents at this time is still very sincere.
"I don't have any doubts about Mr. Li's sincerity or financial resources, I heard Alix say that Mr. Li's bid last year was absolutely overwhelming." This is actually a bluff, Alix naturally would not say this kind of detail to outsiders, but to Li Yanhong's ears, such a detail definitely greatly increases the credibility of the words.
Although he didn't mention more specific details, he said that if it wasn't for Mr. Hinton's back injury, they would have liked to have added whiteness. This was actually revealed in Hinton's own interview later, Meng Fanqi used two details, it seems that he is very familiar with Hinton's team and has competed for a certain initiative.
"You've been connected to someone from Google Brain?" Li Yanhong remembered last year's bidding, and he was still angry, this matter was really unlucky.
Originally, Baidu's first proposal of 10 million dollars had almost impressed Hinton, and Hinton, who had taught all his life, had never seen so much money, and immediately expressed his intention verbally.
As a result, before the official signing was finalized, his two students stopped their mentors and came up with the idea of setting up a shell company for tech giants to bid on and hire through acquisitions.
This comes and goes, and as soon as the time is delayed, it gives Google time and space to operate.
In the final bidding, Robin Li obviously did not lose, but the other three people were interested in Google. Seeing that Google's bid was already higher than its own psychological price, it directly finalized the result.
Do things like this, what else do you bid, go directly to Google to talk?
Although the performance of whiteness on the surface has been very decent in that incident, Li Yanhong is still a little unhappy in his heart.
At the beginning, Google won the Hinton team, and now Yanjing has produced an incredible undergraduate, and the school is still next to his headquarters in Baidu.
Li Yanhong feels that this is the time to operate, and there is a reincarnation in heaven.
Canada is close to you in the United States, is Yanjing closer to you in the United States?
Therefore, Li Yanhong drove to the meeting today, and he was still in a good mood.
But I didn't expect that before I talked a few words, the situation and trend would be the same as last year.
"Google Brain has indeed sent me a letter of intent, and the conditions are quite favorable." Meng Fanqi said calmly, "To be honest, I have basically decided to join, so I said earlier that I would skip the recruitment part in order to save Mr. Li's time." ”
"Mr. Meng, no matter what conditions Google offers you, you can take their contract directly to Baidu." Li Yanhong recalled the situation last year, and stopped bidding back and forth, and directly gave a cap condition.
"Whatever price point you can talk to Google, I'm 20 percent more." Li Yanhong's expression was heavy, and he said word by word.
Last year, for the Hinton team's more than $40 million, it was unable to win, and this year, Google can't offer a higher price for this undergraduate.
Although Li Yanhong said that no matter how much, he actually still has a psychological bottom line, as long as it does not exceed his offer last year, he can accept it.
Meng Fanqi laughed secretly in his heart when he heard this, if he was really courageous, he didn't care how much others bid, and he could only be confident by quoting a price directly.
I have to take other offers to ask for a price, and no matter how I listen to it, it feels a bit lacking in pricing power.
How much is it worth, didn't you even count it? It also needs to look at other people's bids.
However, since Meng Fanqi didn't plan to join, he didn't think much about it.
"Mr. Li, of course the matter of money is very important, but it can't be said to be the whole thing. I haven't fully negotiated the price with Google, and the price they're offering me is about a million dollars. ”
"I have a lot of other factors that I care about, such as my studies, and you may know that I am on a sabbatical, but it's not a way to take a break all the time, I have to find a university like Stanford."
"Of course, these are actually my personal affairs, and they are small things. The more important thing is whether the current situation and atmosphere of Huaguo are really suitable for the development of cutting-edge AI technology. ”
Although Meng Fanqi has the advantage of being reborn, there is an elder who said it well.
People, you don't know, you can't predict it.
A person's destiny is, of course, dependent on self-struggle, but it is also necessary to take into account the course of history.
"Is the atmosphere in China suitable for the most cutting-edge AI technology development? What do you mean by that? Li Yanhong is now also on the road of pioneering, and this matter is naturally something he is very concerned about.
"Li is always the only person who is willing to come to me in person at this time, so I will also pour out my heart and lungs to tell you a few words from my heart." Meng Fanqi reviewed the overall development of AI and said, "AlexNet and my DreamNet have now completely opened the door to deep learning, and it is foreseeable that in a short period of time, there will be a huge breakthrough in the performance of AI models in all directions and fields." ”
"Yes, now is the time to compete for speed, one step is faster, one step is faster." Robin Li nodded yes, affirming this premise.
"Technical indicators are one thing, and from the overall trend, if you want to really implement AI technology and make it available to everyone, it is nothing more than making the model bigger or smaller."
"To be big or to be small...? Do you mean terminal or cloud? Li Yanhong quickly followed Meng Fanqi's train of thought.
"That's right, academic research doesn't pay much attention to the size of the model's parameters and its computational loss at first. However, when it is actually put into use, it will inevitably be divided into a small model with extreme optimization and reduction, and a large cloud model that emphasizes the performance of open interfaces for people to use. ”
"In the early days, the structure and design of the model may be changed and adjusted repeatedly to make the model as fast and as good as possible. But over time, with the accumulation of data, filtering, higher quality annotation, and most importantly, a larger number of model parameters. ”
"Extremely large models will completely produce qualitative changes at a certain critical point, making most small models lose their value. In particular, language speech classes, and image generation class models. The experience of using them is not downward. ”
Meng Fanqi recalled that before his rebirth, Mr. Li faced ChatGPT (version 3.5) and GPT-4, and forcibly resisted the pressure to launch the large model Wenxin early, and had to sigh in his heart.
In fact, Wenxin Yiyan is not useless, but this kind of intelligent model that interacts with people, if dwarfed, will seem particularly vulnerable.