It is actually very difficult to build an AI large language model platform.
Although since ChatGPT's shocking launch, China has quickly followed suit, and several similar language models have appeared one after another.
Even if the requirements are relaxed and all companies trying to grab market share are included, the situation in China alone will reach the level of a "hundred-model war."
There are all kinds of "big models", no matter they exist independently or are sold under the guise of sheep meat, in short, if you add together all the "big models" that look decent, the total number is even more than one hundred and thirty!
Among them was an experimental product that Zheng Hongyi secretly made.
Limited by the size of funds and the company's continuous decline in recent years, Zheng Hongyi did not dare to enter this new track in a high-profile manner.
They are afraid that if they are not careful, if the development is not good, it will be a heavy blow to the company that is already in poor condition.
If the stock price plummets at that time, it will be a big joke.
But even though it was just a secret experiment, the difficulties encountered in the process still gave Zheng Hongyi a headache.
The most troublesome issue is the cost!
The initial cost of simply building a model is not high. It costs about 10 to 15 million to roughly build a large model platform.
However, based on the hardware and training requirements of large models, it is obvious that as the scale of the model grows, the cost of training will become more and more expensive!
First of all, the most basic cost is data!
Large language models are trained on large-scale datasets containing huge amounts of data. In addition to open source datasets, any company that wants to train a qualified large language model must have various datasets accumulated over a long period of time.
Furthermore, data costs also involve data cleaning, labeling, organization, and storage. Both management costs and processing costs are beyond the reach of ordinary companies.
What makes Zheng Hongyi even more worried is that if he wants to build a large language model, the hardware cost is also astronomical!
The GPU, or microprocessor, that is, the one currently most commonly used in large language models is the NVIDIA A100.
Due to the huge demand for GPUs for large language models, the price of NVIDIA A100 has skyrocketed, from the official price of $15,000 to $30,000, and even so, it is difficult to buy it.
Because A100 has been banned from sale...
This is just the price of a single piece!
According to ChatGPT's current size, to meet ChatGPT's needs, at least 30,000 to 40,000 GPUs are needed...
If the user base continues to expand, ultimately reaching the goal of supporting global user access, the number of GPUs required will have to increase by at least a hundredfold, reaching over four million...
The use of hardware on such a scale means that the daily electricity consumption alone is enough to give someone a myocardial infarction!
Originally, after figuring out the cost requirements of the relevant large language models, Zheng Hongyi had already planned to give up competing in this field.
Considering the current situation of 365 Company and its attractiveness to capital, if one attempts to forcibly intervene in this track which is like a financial black hole, the only result will be to become cannon fodder.
But... just when he was about to give up, he discovered the 'Boundless Dimension' platform!
When Zheng Hongyi first discovered the "Boundless Dimension" platform, he didn't take it seriously. He thought it was just like many platforms created by small companies, which was a wolf in sheep's clothing.
But after using it briefly, Zheng Hongyi keenly realized that this 'boundless dimension' platform was completely different from what he imagined!
Then he used the platform to do some practical work, such as asking the platform to make a PPT, or to produce a detailed manuscript based on the specific requirements he put forward, and so on.
After repeated trials, Zheng Hongyi was 100% sure that this "Unbounded Dimension" platform, which appeared out of nowhere, was not only much better than those large language models in China, but was also at least one version ahead of ChatGPT, which is the world leader!
And most importantly, the 'Boundless Dimension' platform is far more in line with Chinese people's usage habits than ChatGPT!
With such a judgment, Zheng Hongyi was so excited that he could hardly contain himself!
Then he used his connections to look up the details of the 'Boundless Dimension' platform.
He was immediately surprised to discover that the company operating the 'Boundless Dimension' platform had only just been established, and the 'Boundless Dimension' platform was the company's first product!
The exaggeration is that the registered capital of that company is very low, and the official statistics show that there is currently no one insured person!
No matter how you look at it, this is a company that can’t be any smaller!
Zheng Hongyi really found it difficult to understand how a small company that seemed to have been established specifically to defraud subsidies could have developed the "Boundless Dimension" platform.
But this is not important!
The platform itself is already very mature. In Zheng Hongyi's view, it can be directly put into commercial operations.
That means that if he can acquire this company, he can save an unknown amount of money just on the initial R&D and construction costs of the platform!
This kind of thing is actually quite common in the Internet field.
When some talented engineers are inspired, they are often able to develop products that are ahead of the times simply by relying on their own talents.
But after developing the product, when it comes to product maintenance, promotion, and commercial operation, these talented engineers will be at a loss and become food for the big sharks!
Zheng Hongyi felt that his luck had changed!
Through investigation of this company, Zheng Hongyi found that there are only two people in this company, both of whom are very young.
One of the people who held more shares looked like the talented engineer he knew, while the other young girl who only held 10% of the shares had a very good family background.
Such a combination is extremely normal in Zheng Hongyi's eyes, and it also makes him more confident in acquiring this company!
It is obvious that a company like this, which is established based on inspiration, will inevitably face the problem of insufficient funds!
When it comes to large language model platforms, the problem of lack of funds will only become more serious!
Taking advantage of the fact that this company was just established less than half a month ago and is in the blind spot of several other domestic Internet giants and has not yet been noticed by those giants, Zheng Hongyi is determined to acquire this company!
With these thoughts in his mind, Zheng Hongyi found that he had arrived.
After calming himself down and returning to his usual indifferent attitude, Zheng Hongyi got out of the car.
But then, Zheng Hongyi froze in place.
The office building in front of me is empty...
Who greeted him?!
Continue read on readnovelmtl.com
