How the quality of training data affects language models
Machine-assisted translation of the archived article. Read the Chinese original.

One of the evolutionary paths of the big AI model can come from a constant model.
By improving the quality of training data,
Raise the effect of the model,
This high-quality data set,
It's a big decision about the quality of the large model.
But the problem with them is that these data come from reptiles in large quantities.
It's the first time I've been able to get to the public.
First of all, copyright is a big problem.
The second is that the content without copyright may not be of high quality.
In the future, high-quality content may be subject to payment.
A large overseas model for training data sets,
An example of this is OpenAI.
The content of the public, free of charge, comes mainly from encyclopedia books and periodicals.
These things are of high quality.
There are no charges at the moment.
But there will be doubt as to whether the fees will be paid.
Another part of the training data for OpenAI, and comon
It's like a webshot of this, and it's like that.
The data set and so on is also very important for the ChatGPT.
Because in these resources you can see human emotions.
For example, there are a few good points.
And their very existence must be cost-effective.
The higher the quality or scarcity, the greater the quality.
The more expensive they are in fact, the higher they are.
And they're probably going to charge for it soon.
In April 2023, the company that trains ani chat robots charged a fee for data use.
Now the time has come for us to see each other.
How to obtain high-quality, low-cost data that do not violate rights,
It may even be the key to future success or failure.
The data may become the age of artificial intelligence.
The most important building block outside of hardware is that it is not the only one.
The value of data has finally become apparent in the foreseeable future.
And the very interesting thing is,
It's about data finance.
This is a government-owned source of income based on ownership of data assets.
On the one hand, the government itself has a lot of data.
That's what our big home model needs.
And just as land has driven the local economy,
The exploration of data,
It is likely that this will also improve the quality of economic development.
On May 14, 2023, the news was broadcast.
There's a line like that.
Chongqing's authorities are breaking down the data barriers, and they're not going to be able to do anything about it.
It's a data-sharing search.
Improved use of data assets,
And we can think about it a little bit.
The future may be an age of artificial intelligence.
The data is a big artificial intelligence model.
Even the key to competition between countries.
I don't care what the government's own data means.
It's possible that the official trading platform will be used for this purpose.
It's a platform for security and quality to trade.
The big data of the corporate media.
It is possible that the development of artificial intelligence itself will be driven more than ever before.
and the productivity of society as a whole,
Even become part of the competition of big countries.
