Beijing-based Baichuan Intelligent Unveils New Large Model “53B”
Beijing, China – August 8, 2022 – Baichuan Intelligent, the emerging leader in the artificial intelligence (AI) industry, unveiled its latest large model, “53B,” on Monday. Following the successful release of Baichuan-7B and Baichuan-13B a few months ago, the company’s third large model aims to further expand their product matrix.
Baichuan-7B and Baichuan-13B garnered positive reviews and were widely adopted by numerous companies. According to Wang Xiaochuan, CEO of Baichuan Intelligent, over 150 companies have already applied to use the company’s open-source models. This success motivated the development of the new large model, Baichuan 53B.
Baichuan 53B incorporates the knowledge and experience gained from the previous two models. The model boasts high-quality, diverse, and hierarchical pre-training data, which significantly enhance its training capabilities. Moreover, Baichuan has leveraged the support of Sogou’s genes to integrate search capabilities into the large model. This addition empowers users to optimize their search experience, improving intention understanding and user prompts.
Baichuan has also implemented alignment adjustment, multi-objective optimization, and multi-RM fusion technologies into the 53B model. These advancements enhance the model’s effectiveness and reliability, ensuring better performance and accuracy.
However, the most significant change with the new model is Baichuan’s decision to make it closed source. Unlike the previous two models, which were open-source, Baichuan 53B will be proprietary. Wang Xiaochuan explained that the larger scale and higher deployment cost of the model necessitated this change. The company believes that a closed-source approach will provide a simpler interface and greater convenience for customers compared to an open-source model.
When asked about the shift from open source to closed source, Wang Xiaochuan emphasized the variety of opportunities in today’s market. He revealed that Baichuan’s future plans extend beyond small business operations and consumer-facing products. The company aims to explore the potential of To B services and other business opportunities.
Earlier this year, Wang Xiaochuan announced that Baichuan would release a large model with a parameter level exceeding 50 billion by Q3 2022. Surpassing expectations, the company has achieved this milestone ahead of schedule. This achievement raises speculation about whether Baichuan can fulfill its promise to benchmark against ChatGPT3.5 by the end of the year, as mentioned in April.
Baichuan’s strategy execution and funding have also been successful. The company recently completed its second round of financing, achieving a valuation of approximately $500 million. With these accomplishments, Baichuan is making steady progress towards becoming a unicorn in the large-model AI industry.
Leifeng.com recently conducted an in-depth interview with Wang Xiaochuan to delve into his entrepreneurial journey over the past six months. The interview covers topics like team recruitment, building a successful AI company, and Wang Xiaochuan’s insights into the future of breaking latest news (Artificial General Intelligence) business. The full article will be released soon on Leifeng.com.
Readers interested in discussing AI and exchanging knowledge are also welcome to connect with the author, William Dong, on WeChat (account: william_dong).
Please note that unauthorized reproduction of this article from Leifeng.com is strictly prohibited. Refer to the reprint instructions for further details.