Tencent is about to miss out on the "Hundred Model Battle" of the Big Model Era

The 6th World Artificial Intelligence Conference was held in Shanghai from July 6 to July 8, 2023.The National AI Standardization Team announced that the team leader of China's big model standardization theme group will be jointly served by Shanghai AI Lab, Huawei, Baidu, Ali and other enterprises in the future

The 6th World Artificial Intelligence Conference was held in Shanghai from July 6 to July 8, 2023.

The National AI Standardization Team announced that the team leader of China's big model standardization theme group will be jointly served by Shanghai AI Lab, Huawei, Baidu, Ali and other enterprises in the future. In these enterprises, after careful searching, there is no shadow of Tencent. Everyone knows that companies that can take the lead in drafting standardization are leaders in new industries, and it is obvious that Tencent and its 17 ecological partners are not at the forefront.

Some netizens online say that Tencent Games makes too much money, and some department leaders within Tencent are no longer willing to develop new technologies and fields with high investment, slow returns, long cycles, and low profits. After all, releasing a new skin can sell for billions of yuan in a day when it's hot. The profits this brings to the company are real, the department's performance is good, and the department leader can still be promoted. Why not do it!

A few days ago, Huawei's Hu Houkun said at the meeting, "Currently, half of China's large model innovation is supported by Ascension AI.

Developing large models is expensive and technology intensive. Huawei is willing to do it, but Tencent is unwilling. Tencent likes to imitate.

We just don't know that in the field of artificial intelligence, there has already been a "hundred model war".

Firstly, how to understand a large model and give you an example:

You can compare various development companies to universities. The top product of universities is doctoral students, while the top product of companies developing artificial intelligence is big models. This doctoral student has knowledge of astronomy and geography, and can know both the past and the future. The "Hundred Model Battle" is a comprehensive competition among doctoral students graduated from various universities. Whoever wins represents which university's technological strength is more advanced, and whoever wins represents which company's research and development of large models is more intelligent.

Doctoral students receive training and training from birth, and will input all the knowledge known to humans, including Wikipedia, web pages, ancient and modern books, to train the intelligence of large models.

Since the "Hundred Regiments Offensive", many people, individuals and enterprises rely on the group buying platform to work, earn money and live daily. It can be foreseen that after the "Hundred Model War", there will also be many people, individuals, and enterprises relying on large models for work and life, including every one they are working on, unable to avoid and avoid.

I am pleased to see Baidu participating in the standardization of China's large model this time. Originally, Baidu was a company with great potential for development and was highly worth investing in. However, during its development process, especially after Google withdrew from the Chinese market, when Baidu emerged as the dominant company in the search field, it forgot its original intention and experienced a series of events such as bidding rankings and Weijie. As a result, Baidu's position as the leader in the search field was not guaranteed, and its market share was gradually being plundered by peers, gradually losing ground, and being abandoned by users, There is a great possibility of exiting the historical stage.

Thanks to Baidu's ability to brake in time and devote all its energy to artificial intelligence, automatic driving and Big data, it can continue to take the lead in the new round of technological revolution. I hope Baidu can not forget its original intention, draw on the mistakes and experiences made in the previous stage, and shine again in the new big model AI field.

The large model has gone through three stages of parameter size growth: pre training model, large-scale pre training model, and ultra large scale pre training model. The parameter size has increased from the initial level of one billion to the current level of one million billion.

The development process of AI large models (the larger the pre training parameters, the more intelligent the large model):

In 2018, Google first proposed a large-scale pre trained language model: BERT, with pre trained parameters exceeding 300 million in scale;

In the same year, OpenAI proposed the pre training model GPT;

In 2019, OpenAI launched GPT-2 (second generation) with pre training parameters of 1.5 billion;

Nvidia launched Megatron LM with pre training parameters of 8.3 billion;

Google launches T5 with pre training parameters of 11 billion;

Microsoft launches Turing Turing NLG with pre trained parameters of 17 billion;

In 2020, OpenAI launched GPT-3 (the third generation), with pre training parameters reaching 175 billion, the first large model exceeding 100 billion;

Microsoft and Nvidia jointly released Megatron Turing (MT-NLG), with pre training parameters of 530 billion;

In 2021, Google launched the SwitchTransformer model, with pre trained parameters of up to 1.6 trillion yuan, making it the first trillions level language model for humans;

SenseTime releasedINTERNPre training parameters of 10 billion;

Huawei Cloud and Circular Intelligence Joint ReleasePangu NLPModel with pre trained parameters of 100 billion;

Huawei Cloud and Peking University jointly releasedPanguPre training parameters of 200 billion;

Alibaba Dharma Institute released PLUG model with pre trained parameters of 27 billion;

M6Model with pre trained parameters of 100 billion;

ERNIE 3.0Pre training parameters of 10 billion;

Inspur LaunchSource 1.0 ModelPre training parameters of 250 billion;

Baidu has launched the ERNIE3.0Tian model again, with pre trained parameters of 260 billion;

In 2022, Alibaba Dharma Academy and Tsinghua University were establishedBAGUALU modelThe pre training parameters are 174 trillion yuan.

According to public data, the number of large models released by domestic enterprises and research institutes has exceeded 80, ranking second globally.

Ma Huateng, CEO of Tencent, also mentioned that the AI model is a once-in-a-century historical opportunity similar to the industrial revolution. But so far, Tencent has not released any real big model products, only indicating that Tencent's industry big model has achieved several times the efficiency improvement in financial risk control, interactive translation, and customer service scenarios such as Zhiren.

However, Tencent Cloud launched the MaaS (Model as a Service) one-stop industry model service on June 19th, providing solutions for customers to implement large models.

I hope Tencent can keep up with the pace of the times and, with its technical talents and financial strength, launch its own large-scale model products as soon as possible, making more contributions to society and the country.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])