Kai-Fu Lee (Photo credit: TMTPost App editor Lin Zhijia)
AsianFin -- 01.AI, the AI unicorn founded by renowned tech entrepreneur Kai-Fu Lee, announced on Monday the launch of its "WanZhi Enterprise Large Model One-Stop Platform."
Designed to provide customized enterprise-level DeepSeek deployment solutions, the platform aims to accelerate the adoption of large models in various industries.
The WanZhi platform comes pre-equipped with high-performance GPUs, including options like Ascend, and integrates the full range of DeepSeek models, such as V3 and R1. It supports cutting-edge model architectures like DeepSeek, Qwen, and Yi, offering flexible scalability and internet search capabilities.
The platform's AI search product boasts an impressive accuracy rate of 88%, effectively correcting large model information and outperforming other mainstream AI search solutions.
01.AI has developed a highly scalable RAG architecture that supports both structured and unstructured data, doubling deployment and implementation speeds. Advanced Rewrite and Rerank models further reduce hallucinations in large models like DeepSeek, improving recall rates by 60% and accuracy rates by 30%.
In an exclusive interview with AsianFin, Lee stressed the significance of the "DeepSeek moment," saying "DeepSeek is not only an excellent model, but also an open-source one, which has a huge impact internationally."
The WanZhi platform enables any enterprise to quickly adopt DeepSeek and continuously improve its performance over time." Lee highlighted that the platform addresses enterprise needs in secure deployment, practical applications, and industry-specific customization.
Lee also shared his vision for the future of large models, predicting that 2025 will be a breakout year for AI-First applications and a critical period for the commercialization of large models. "The landscape of large models will shift from competing on foundational model technical metrics to embracing open-source models for commercial empowerment," he said. "China, with its massive market and abundance of application scenarios, is well-positioned to lead this transformation."
01.AI, founded in May 2023, initially focused on self-developed large models. However, the company has since shifted its strategy from a "model + application" approach to an "application + model" approach. This shift involves using application scenario demands to reverse-engineer the capabilities of industrial large models, bridging the gap between foundational models and vertical scenarios.
The company's decision to cease iterations on ultra-large parameter foundational models and focus on lightweight, commercially viable models like Yi-Lightning has proven successful. Yi-Lightning achieved a world ranking of sixth in the LMSYS Chatbot Arena in October 2023, demonstrating the effectiveness of this strategy.
On the commercialization front, 01.AI has adopted a dual strategy, targeting overseas consumer markets (C-end) and domestic enterprise markets (B-end). Last year, the company generated over 100 million yuan in overseas revenue, and its domestic B-end revenue in the first quarter of 2024 has already approached the total revenue for the entire previous year.
The WanZhi Enterprise Large Model One-Stop Platform is designed to lower the technical barriers for enterprises while ensuring the security of localized data deployment. It primarily targets small and medium-sized enterprises (SMEs) with standardized products that allow for rapid delivery. For larger, customized orders, non-standard solutions are also available.
01.AI believes that the platform's comprehensive capabilities in "ultra-large-scale computing power management + data asset and security management + industry model training/fine-tuning + industry application platforms" make it a powerful tool for enterprises. The company plans to integrate multimodal capabilities and digital human services into the WanZhi Platform in the future.
Lee reiterated 01.AI's mission to "Make AI Work" and emphasized the importance of respecting business principles. "The future competition in the large model industry will no longer solely focus on model performance but will also hinge on the ability to transition from the middle platform to applications," he said. "This means whether the model can quickly respond to scenario demands and build industry applications based on the middle platform."
更多精彩内容，关注钛媒体微信号 （ID：taimeiti），或者下载钛媒体 App
