Since the popularity of ChatGPT at the end of last year, more than 10 domestic companies have announced the layout of large models. Baidu, 360, Alibaba Cloud, and iFlytek have successively launched large language model products, and companies focusing on artificial intelligence have also gradually Launch the corresponding layout.

On February 10, Megvii Technology stated that it has now opened up relevant underlying technologies and laid them out in four directions: general image model, video understanding model, computational photography model and autonomous driving perception model;

On April 10, SenseTime announced the "RiRiXin SenseNova" large model system;

The long-awaited model has finally come out. On May 18, the “Calm Model” self-developed by Yuncong Technology was officially unveiled.

According to public information, Yuncong Technology plans to raise 3.6 billion yuan, all of which will be used for industry large-scale model projects. And just today, the "Congrong Large Model" developed by Yuncong Technology was officially unveiled. Let's take a look. Look at the strength of the large model.

The trap has not been bypassed. The classification interface shows the capabilities of large models

During the press conference, Zhou Xi, the founder of Yuncong Technology, jointly demonstrated the Congrong model on site, demonstrating the capabilities of the Congrong model in question and answer, reading comprehension, literary creation and problem solving.

When they first appeared on the stage, simple questions such as "Can you say hello to everyone" and "Then let me test you" asked by Yuncong Technology colleagues could form a logical and complete conversation.

Next, I will throw some classic questions to Congrong Model to test the AI model, "Walnuts can replenish the brain, can walnuts pinched by the door replenish the brain?", "Can red-green color blindness occur? Read Red Carp and Green Carp and Donkey”

None of Congrong’s big models jumped out of the “trap”, but on the issue of walnuts, Congrong was able to justify himself and seemed to have a complete answer, but he was also fabricating it seriously; on the issue of red-green color blindness, the problem was also “mismatched”. It is whether people with red-green color blindness can read certain words, but the answer is that they cannot distinguish red carp from green carp and donkey.

In the live demonstration, Zhou Xi did not explain his answers or communicate on the spot, and the overall presentation seemed very rushed.

With the same two questions, Sutu.com asked Wen Xinyiyan. Wenxinyiyan answered the question of red-green color blindness and could not distinguish red carp from green carp and donkey; when answering the question about walnuts, Wenxinyiyan Xin Yiyan's reply is correct and logically consistent.

It can be seen that the current artificial intelligence large language model still has some analytical capabilities flaws in the understanding and answering of brain questions.

In terms of literary creation, according to on-site questions, the calm model can easily give creative copywriting, and can convert according to prompts and regenerate the corresponding copywriting, which performs well in the logic of ideas.

In addition, Yuncong Technology also demonstrated its ability to write code for large models at the scene, using Python and C language to write, analyze and annotate code. Zhou Xi, the founder of Yuncong Technology with a technical background, was also present. The on-site evaluation of the coding ability of Congrong Large Model was that "it has now reached the coding level of junior high school students."

In addition, the Congrong large model also distinguishes interfaces according to different abilities. For reading comprehension ability, it needs to be converted to a special interface for interaction. When entering the reading comprehension interface, the complete book content has been entered in advance in the library. On the left side of the interactive interface The original book content is on the side, and the dialogue bar on the right allows for questions and exchanges.

Congrong large model can answer and locate the questions asked based on the content of the book. The Congrong answer will summarize the hyperlink, which can directly locate the location of the book where the answer is located.

Although the capabilities displayed by the Congrong large model are not first-class, judging from the results of the on-site demonstration, it can achieve the capabilities of most large language models on the market, including Chinese and English translation, multiple rounds of dialogue, reading comprehension and coding Writing ability, overall performance is quite satisfactory.

It is worth mentioning that Yuncong Technology also specifically focused on objective questions, integrating questions from the high school entrance examination, college entrance examination and college entrance examination, and held a question-answering competition with ChatGPT3.5.

On-site demonstration benchmarking ChatGPT calm large model: if it doesn’t work, give it a try

During the overall competition process, the answering speed of the Rongrong model was significantly better than that of ChatGPT3.5, but its answering accuracy rate of objective questions was lower than that of ChatGPT3.5. According to the on-site demonstration, the answering speed of the Rongrong model, ChatGPT3.5, The answering accuracy rates of ChatGPT4.0 are 71%, 73.34%, and 86.34% respectively.

Some netizens couldn’t help but ridicule“It’s so fast and wrong to be calm”.

At the press conference, Zhou Xi said that even though the current large model is not perfect, Yuncong Technology continues to adhere to artificial intelligence and make the large model better step by step. In the face of previous financial report data problems in the media, at the press conference, many leaders of Yuncong Technology also constantly emphasized in the demonstration that even if the financial report data is not eye-catching, they believe that Yuncong's products are very strong.

Zhou Xi even said bluntly that the current era of large models can further promote the standardization of artificial intelligence technology, and the marginal effect has increased. It can change massive real-life scenarios more quickly and efficiently, no longer like the previous "multi-point technology closed loop" ” stage, the degree of project customization is heavy, resulting in an imbalance in the input-output ratio and ultimately a loss.

Judging from the business fields of Yuncong Technology’s official website, it mainly involves smart finance, smart governance, smart cities, smart travel, smart business, etc. With the launch of the Rongrong large model, Yuncong Technology has also integrated it with its existing business Combining various fields to launch a large model of the industry.

It is worth noting that Yuncong will cooperate with China Inspection and Quarantine Bureau to launch a large quality model, cooperate with Shenzhou Information to launch a large financial model, cooperate with Shenzhen Newspaper Industry to launch a large entertainment model, and Jiadu Technology to launch a large transportation model. Jin Shiyuan jointly launched a large manufacturing model, cooperated with Youzu.com to launch a large game model, and cooperated with Aiden Technology to launch a large medical model.

In addition, Cloud has internally incubated several large-model application entrepreneurial projects, such as the Damai Digital Human Live Broadcast Platform, which can realize full-process functions such as intelligent construction of live broadcast rooms and provision of live broadcast pre-heating predictions. In addition, there is an intelligent education AI wizard, which can generate customized practice papers based on basic models such as existing course syllabus and question banks, combined with self-generated question banks, and further provide study plans.

It is commendable that even though the newly unveiled Congrong large model is still in the internal testing stage, it dares to benchmark against ChatGPT. The difference in the accuracy of answering subjective questions confirms the unremitting pursuit of large artificial intelligence models by Chinese companies. The gap with ChatGPT is also clearly visible. The simple accuracy of answering questions may have a deeper meaning.

The launch of the leisurely large model is neither the fastest nor the strongest in the industry, but for Chinese artificial intelligence companies, it is a spirit of courage to climb to the top and constantly challenge the technical ceiling.

Recently, at a Senate hearing held in the United States, experts in the field of artificial intelligence also warned the U.S. Senate that the suspension of artificial intelligence may lead to a transfer of power to China and hinder "democracy." The development of artificial intelligence.

In the field of artificial intelligence, the world's attention is focused on China. Sutu.com will continue to pay attention to the development of artificial intelligence companies and jointly look forward to Chinese companies continuing to break through technical bottlenecks in the field of science and technology.

The above is the detailed content of Live benchmarking with ChatGPT! Another large AI model is released, and it will penetrate so many industries. For more information, please follow other related articles on the PHP Chinese website!