>首页> IT >

天天观焦点:Zhihu Debuts GPT-4 Rival Model, Jumping into Tech Giants' AI Race

时间:2023-04-14 10:18:58       来源:钛媒体


(资料图)

BEIJING, April 13 (TMTPOST)— The Quora-like Chinese online content website Zhihu rushed into the artificial intelligence (AI) race that is heating up despite increasing regulation.

Source: Visual China

Zhihu unveiled Zhihaitu AI, a large language model (LLM) co-developed with Model Best, a Beijing startup engaged on AI models. Zhou Yuan, the founder and CEO of Zhihu, announced to start the internal testing of the first LLM of his company on Thursday. Test is set to center on Zhihu Hot List, which will function as a source for the model to fetch, organize and aggregate the useful content to reply users for their answers with an outline.

"During the testing, the LLM developed by Zhihu and Model Best under the scenario of Zhihu Hot List has delivered the same performance as GPT-4," Zeng Guoyang, the cofounder and Chief Technology Officer (CTO) of Model Best, compared Zhihaitu AI with the latest version of AI model developed by OpenAI, the company behind ChatGPT.  In the future, Zhihu will expand available scenarios for AI LLM application, Zeng said at a release event Thursday. Zhihaitu AI was trained on Model Best’s CPM-Bee model. The model is the better than any other LLMs to process Chinese language based on our evaluation, Li Dahai, partner and CTO of Zhihu, introduced. He added all the founding members of the model came from the Natural Language Processing and Computational Social Science Lab, Tsinghua University.

Zhihu’s launch came days after three Chinese tech giants intensify the AI race. Alibaba announced on Tuesday to integrate its own LLM Tongyi Qianwen into all the company’s offerings and first be deployed on DingTalk, an enterprise communication and collaboration platform, and the smart speaker Tmall Genie. Lark Suite, an enterprise collaboration platform created by TikTok parent ByteDance, was said the same day to roll out an intelligent assistant named My AI, without definite timetable about its public test and launch. The product can help users to automatically summarize meeting minutes, creating reports, improve the text content and complete the unfinished writing; moreover, it is useful for Lark app users for automatic schedule creating and search in the knowledge base inside by means of dialogue.

Tuesday also saw a draft for a new regulation on AI-based generative contents and services, the first of the kind in China since the launch of ChatGPT last November. The Cyberspace Administration of China (CAC) requires companies to offer generative AI services to take steps to avoid generating false information for the content generated by such AI services shall be true and accurate, according to the draft. Providers for ways including programmable interfaces to support generative AI contents shall take responsibility as the creator of the content produced by the product, and they have to protect personal information if their possessions involve individual users’ information.

A day prior to CAC’s new rule, SenseTime, the leading AI software company in China, just rolled out a range of AI-powered tools, including the ChatGPT-like SenseChat. The chatbot, as a 180-billion parameter Chinese LLM, is built on the latest version of the firm’s SenseNova big model, which has been developing for five years. Besides multiple rounds dialogues with users, the bot also can be used to create contents, analyze sentiments, generate images, and script computer codes.

关键词: