news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

how does netease use big models internally? how does it internalize and externalize big model technology products?

text｜niu hui

‍‍editor｜zhao yanqiu‍‍‍

this year, the big model competition has shifted from last year's hundred models competition to industry implementation competition.

according to statistics from digital intelligence frontline and other organizations, the number of large-scale model public bidding projects is increasing month by month.the number of cases increased rapidly from 15 in january to 127 in august, setting a new monthly record.from the perspective of industry distribution, operators, finance, energy, automobiles, internet, education and other industries are active, and enterprises are accelerating the implementation of large models.

in the hot market, major technology companies are accelerating the construction of large model capabilities and market competition.to b brand——netease digital intelligence, and ai is an important strategic direction of netease digital intelligence., promoting the integration of full-line products with ai.

“one day in ai is like one year in the human world.” at the recent netease digital innovation conference on ai 2024, netease vice president and netease digital general manager ruan liang observed that, whether in the company’s internal environment or in the business process,ai has entered a stage of rapid evolution and iteration。

“although ai still faces many problems, my consistent attitude is:we must firmly embrace ai. ai must also pay attention to the most essential emotions of people and make technology warm. "he said that netease digital intelligence will also focus on "ai, standardization, and industry-based"as the core business direction, we integrate advantageous resources and experience to help enterprises build their own ai productivity.

01

how does netease use the big model itself?

when talking about netease digital intelligence using ai technology to serve enterprise customers, people are first curious about how netease itself applies big models. ruan liang also introduced for the first time the progress of netease's application of big models in some business lines.

netease yanxuan has used big models for product design and marketingpreviously, for different categories of products, such as suitcases and cat food, yanxuan product managers had to design product prototypes after determining product features and market groups, and sometimes even ask models to shoot real scenes before finally coming up with design drafts for competition.this cycle usually takes one or two weeks.now, they use aigc to directly generate design drafts.the time is shortened to one or two days. aigc can also generate unlimited marketing materials for delivery to different platforms.

“these scenarios are applied very smoothly within netease."it saves a lot of costs," ruan liang concluded.the key is opportunity cost"in the past, we would have to wait two or three weeks to find out about the failure of products and marketing.now it's verified in two or three days., we can adjust immediately.netease's financial report will reveal the company's relevant profit margins, r&d and art efficiency has been rising. one of the reasons isextensive use of ai。”

“this also creates a new type of worker, who may not be an art designer.as long as you have two skills - aesthetic ability and imagination, one person can defeat thousands of troops. "ruan liang further said.ai brings equal rights to workers, which will have a profound impact on the industry.

netease cloud music introduces chatbithis is aconversational data analysis productspreviously, data analysis required programming and professional skills, which resulted in many people having data but not knowing how to use it. after adopting chatbi, cloud music's non-technical personnel in product, operation, and marketing all have their own intelligent data analysts who can explore various data analyses.achieving data equality to some extentat present, cloud music has obtained data more than 12,000 times through chatbi. if each q&a saves 0.2 man-days, it saves more than 2,000 man-days.

an example of a chatbi conversation, with simulated data.

there is also a product closely related to "customers" called voice of customer (voc).netease digital intelligence released a customer service model called "shanghe" last year to listen to the voice of customers.ruan liang introduced that netease has always attached great importance to customer feedback, which is the company's secret to making products. now with a large model, it can classify and analyze customer voices from all channels such as customer service conversations, work order data, product reviews, questionnaire surveys, social media comments, etc., and output concrete and traceable demand insights.

"we first found some seed customers and analyzed their voices through ai, which gave us some surprises." ruan liang said that ai had gained insights into some key information from customer complaints that had been overlooked before.sometimes there are positive opportunities hidden in negative information, and negative feedback hidden in positive voices.. "ai has unearthed these "hidden meanings".

recent,netease digital intelligence is still exploring, about employees' voicethey use ai to capture some information from massive data to help human resources and finance discover outstanding talents or certain risks in the internal organization.

netease digital intelligence also uses ai agent to serve as digital twins for company employees. “ai agent is becoming an extension of human intelligence and a digital twin.ruan liang introduced that some netease digital intelligence employees are interrupted by many colleagues' inquiries every day, such as asking about customer conditions, relevant data and information. now, the knowledge these employees have access to and the work records of dingtalk and feishu are used to train their own digital avatars with ai.in practice, digital twins can solve 60%-70% of problems, employees can focus more on their work and their anxiety is relieved."

ai agent can break down targets, perform actions, etc. ruan liang said,this is a very good concept after aigc, and it has great potential in the future.。

02

large model technology in "endogenous externalization"

behind these ai applications is the ai ​​technology developed by netease digital intelligence. these technologies integrate the ai ​​technologies developed by netease digital intelligence in its main business areas:customer experience, digital content security, data value realization, software development, ai interactionon accumulation.

for example, in terms of customer experiencein september 2023, netease digital intelligence launchedbusiness and big models for customer servicethe business and big models are also further applied to products such as voice of customer (voc), intelligent quality inspection, customer service robots, intelligent outbound calls, and precise insights. nearly a year after its launch, these aigc products have served industries such as footwear, automobiles, home appliances, games, enterprise services, and manufacturing.

on digital content securityin july this year, netease digital intelligence releasedthe first large-scale vertical security model in china focusing on digital content risk control scenariosin response to the challenges brought about by the aigc era, such as illusion problems, knowledge ambiguity, and false news, we use big models to understand the risk control scale and improve the efficiency of manual review.the most advanced technology has two sides, security is extremely important." ruan liang added that this is also the responsibility of technology companies.

in addition to these two vertical big models, netease digital intelligence is also promoting the integration of ai with bi, low-code and audio and video technologies, and has launched chatbi copilot, ai+low-code development, and one-stop audio and video solutions based on big models.

after netease started using these ai-enabled products internally, it began to promote and implement them externally.

for example, in the retail industry,cdfg, which has 35 million members, applied the voice of customer (voc) supported by the business and big data model., to analyze after-sales conversations on different channels, find out what users are concerned about in big promotions, and provide targeted “active services”.

netease digital intelligence observed that voc, which is supported by large models, is welcomed by enterprises. the ceo of a gaming company told ruan liang that he hopes to get real feedback and insights. "every time people report to me, they may have whitewashed things. i actually want to see how players 'scold' us." in fact, when brands become bigger and bigger, they often ignore the connection with customers, and successful companies cannot do without good brands. ai technologies, including voc, play a key role in this regard.

this has been confirmed by customers. at a recent customer conference, a representative of a well-known electronics company said that by using the above products, the company has the opportunity to hear more of the real voices of consumers, clarify the direction of product and service improvement, and enhance the brand's consumer experience in china.

in terms of safety,netease digital intelligence's products have helped many leading large model manufacturers in the industry, such as minimaxthe business development of netease digital intelligence has also been applied to scenarios such as forum communities of securities companies and player dialogues in online games. now, netease digital intelligence's security vertical big model will gradually be applied to the services of leading big model manufacturers to enhance the risk control capabilities of digital content and use al to "fight" ai.

based on the test data of various scenarios, the big security model has performed well in the recall of some pornographic difficult samples, the recognition rate of illegal advertisements, and the recognition rate of aigc face stylization difficult cases. in addition, the big security model has improved the risk prevention timeliness to the hour level.

in the field of data analysis,shengye thermal power, a company in the thermal power industry, uses netease digital intelligencechatbi's big model + biforce, achieving intelligent viewing and analysis of more than 95% of the problem data in the thermal power industry. moreover, in shengye thermal power, everyone from ordinary employees to management can "find numbers", "compare numbers", "solve numbers" through dialogue, which has improved the overall operational capabilities of the group.

netease digital intelligence's ai agent has also been applied in automotive pre-sales consultation and e-commerce pre-sales and after-sales customer service scenarios.

03

how can enterprises better implement ai?

in this conference, corporate audiences were very active, and some presentation ppts were "collectively" photographed. companies were concerned about how big models could move from technology to business production, and implementation was also accelerating.

hu kai, general manager of intel's internet industry in china, said thatby 2026, 80% of companies will use generative ai; enterprise spending on generative ai will also increase from$40 billion this year, rising to $151 billion in 2027ianalysis founder and ceo jin jianhua observed that among the companies surveyed, two-thirds have adopted the big model, while in august and september last year, more than 50% of the companies were in a wait-and-see learning stage.

there is a saying in the industry that since large models can directly affect business,has a "oppression effect"once a peer company has landed and achieved results, other companies must follow suit. netease digital intelligence has also felt the enthusiasm of companies to embrace ai. after the overall technology and product upgrades, many companies such as casio, bank of beijing, and bank of hangzhou have reached cooperation with them. ai has been deeply empowered in retail, entertainment and social networking, energy, finance and other industries.

however, the industry has also seen thatthe big model faces a lot of resistance in both capacity building and application promotion. many employees are worried that their jobs will be replaced.this is an issue worth exploring.

jin jianhua of ianalysis suggested that the business department should be involved in the early stage of project construction to find the pain points in their actual business. in the business benefit evaluation, the business team and management must participate.the key is to align the consensus between business management and it department。

in the process of promoting the application of large models in enterprises, jin jianhua introduced that the public support of senior leaders for benchmark projects, the discovery of activists, the regular disclosure of the operation and benefits of benchmark projects and the public incentives have achieved good results in practice.the "work points" system will give extra points based on performance when using ai tools, and it also has a driving effect in actual operations.

from the perspective of the products of the large-scale landing model, according to the research of iresearch,data analysis and knowledge base are the two major tools for enterprises at present., they involve many scenarios. in addition, ai agent is rapidly popularizing and penetrating. from the needs of business departments and functional departments,sales and marketing are still the scenarios that everyone cares about mostit is worth noting thatr&d scenarios also have a high demand for large models, exceeded expectations.

judging from the practice of netease digital intelligence's internal promotion and external technical service provision, positive results have been achieved in the construction and application of large models. "as we actively embrace ai, we will find that ai is not here to subvert us, but to empower us. it can become our wings and let us fly higher," said ruan liang.

therefore, in the tide of big model implementation, netease digital intelligence proposed "ai, standardization, and industry-based"core business direction. among them,aiwe will actively use various ai tools to make products more beautiful and more efficient, use ai tools to solve heavy, repetitive and boring work, and better exert people's creativity and innovation; firmly promote the development of existing productsstandardization; industryit focuses on the gaming, finance, retail, social and entertainment industries to further consolidate and expand its leadership in the industry.

through these "three transformations", the integrated and upgraded netease digital intelligence will devote all its efforts to competing in the enterprise-level ai customer market.