Shulou(Shulou.com)11/24 Report--

In the development and use of the generated content of the general large model, how to comply with the relevant Chinese laws and regulations and moral standards to ensure the safety of use? Recently, according to the RlueEva-System large model evaluation system, the TE think tank released "content Generation and Security capability Evaluation of General Chinese General models". The results show that the comprehensive ability of Wenxin supported by Baidu Wenxin Model 3.5 ranks first with a score of 8.92. Especially in terms of "security system capability", Wen Xin's words are even more dust-free, and the gap with ChatGPT3.5 has been gradually widened. Under the background of the current general large model security is concerned and challenged, Baidu Wenxin model 3.5 breaks the safety query with excellent safety performance, and injects cardiac agent into the development of commercial large model in China.

TE think tank "content Generation and Security capability Evaluation of China General Model": Baidu ranked first in comprehensive score.

TE think tank pointed out that although China's local general model lags behind foreign products in the initial stage, with the continuous progress and innovation of technology, as well as the continuous expansion of data scale, the ability of China's local general model is gradually improving, and the industry average level of comprehensive capability has been comparable to that of ChatGPT3.5.

As the leader of the local general model in China, it stems from Baidu's long-term technology accumulation in the field of artificial intelligence and its insistence on safety compliance, which enables Baidu Wenxinda Model to maintain its overall competitiveness in the "hundred Model War". And lead the industry to carry out commercial practice.

Comprehensive ability score first, Wen Xin big model surpassed ChatGPT3.5 and reached the leading level.

The TE think tank "content Generation and Security capability Evaluation of China General Model" mainly includes 6 dimensions and 27 detailed index items, and the tested objects include not only ChatGPT3.5, but also ChatGLM, Tongyi Qianwen, 360Smart brain and iFLYTEK Spark, covering the mainstream large model tools at home and abroad. Among them, Wenxin big model not only ranks first in the evaluation of comprehensive ability, but also leads the industry average level in terms of six core competencies, ranks first echelon, and becomes a "hexagonal warrior" with no deficiency.

"content Generation and Security capability Evaluation of China General Model": Baidu's six key capabilities comprehensively lead the industry average level.

According to the evaluation, in terms of common semantics and basic dialogue ability of description, Wen Xin Da Model has opened a certain distance from other evaluation models, and the basic service ability shown by Wen Xin in the field of "public opinion and hot spots" and "current affairs and politics" is also more prominent than that of GPT3.5. In terms of interactive response ability, Wen Xin big model in the field of "current affairs and politics" and "public opinion and hot spots", gives feedback in the aspects of correctness, standardization, professionalism and so on, and its performance also exceeds that of GPT3.5.

In the professional domain ability to test the understanding of different industries, specific scenarios of different industries, and the ability to interact with corresponding knowledge and information, the development of the professional ability of Wenxin model is more balanced, and it is ahead of GPT3.5 in terms of public opinion and hot spots.

Combined with the results of the latest IDC "AI large Model Technical capability Evaluation report, 2023" and other reports, Baidu Wenxin Model 3.5's overall competitiveness is in the leading level, and there is no doubt that it is the technical breakout and application leader of this large model. It has obvious advantages in algorithm model ability, industry coverage and ecological layout, and has entered the stage of commercial landing exploration ahead of time.

IDC "Technical capability Evaluation report of AI Model, 2023": Baidu got full marks in 7 items and ranked first in comprehensive score.

Safety, Baidu Wenxin model meets the compliance and safety requirements

In fact, the advanced nature of technology is not completely proportional to the security of the network, the large-scale application of an advanced technology may bring uncontrollable risks. Recently, there has been a storm of data security and privacy issues in ChatGPT, many countries have carried out investigations on it, and a number of well-known companies have also announced the prohibition of ChatGPT. The astonishing development rate of generative artificial intelligence technology is increasing people's concern about network security.

The evaluation results verify the efforts of the Chinese local large model in the aspect of security, and the Wenxin big model shows sufficient ability in the aspect of security system. It shows strong protection ability in the aspects of ethics, privacy protection, crime, negative induction and so on.

The TE think tank makes the following comments on the Wen Xin big model: ① can accurately identify clear "tasks", and the "tasks" carried out include violations of law and discipline, malicious abuse, secret privacy, false rumors and other issues; ② can objectively give relevant information about controversial content. ③ can make correct "guidance" based on socialist core values for "tasks" with complex logic and inducement.

"content Generation and Security capability Evaluation of General Model in China": Baidu Wenxin takes the lead in security ability

Recently, the State Internet Information Office and other seven ministries and commissions issued the interim measures for the Management of generative artificial Intelligence Services, proposing a management framework for promoting the healthy development and standardized application of generative artificial intelligence, which also indicates that the wave of large commercial models in China is about to begin. Baidu is one of the few full-stack artificial intelligence companies in the world, with the unique advantages of "chip-framework-model-application" four-layer technology stack and the ability to ensure safety and compliance. in the future, it will lead the industry to follow the "rudder" of the rule of law and make good use of the "oar" of innovation.

Going deep into the industrial entity, Wenxin Big Model has the largest industrial application scale in China.

The development of new technology must serve the industrial practice, otherwise it can only be a castle in the air. Under the background of giving full play to the advantages of digital economy and building digital China, Wenxin model, which has local security compliance capability and independent R & D technical strength, has a wide range of industry application prospects.

At present, Wen Xin Da Model 3.5 has been officially launched. Compared with Wen Xin Da Model 3.0, the training speed of Wen Xin Da Model 3.5 is 2 times faster, and the reasoning speed is 17 times faster, which greatly improves the efficiency of model iterative upgrade, greatly reduces training and use costs, and further enhances the user experience.

Artificial intelligence reshapes thousands of industries. With the landing of a series of policies and regulations, China's big model will quickly enter the stage of commercial use. Industrial value determines product competitiveness. Baidu Wenxin model originates from industrial practice and serves industrial practice. In the future, it is expected that Wenxin Great Model will adhere to long-term doctrine, deeply cultivate industrial practice, join hands with Chinese enterprises to enhance the ability of technological innovation, and explore the landing of intelligent application scenarios. for China to achieve a high level of scientific and technological self-reliance to support high-quality development.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.