(Yicai) Feb. 10 -- China's financial industry is stepping up the deployment of artificial intelligence startup DeepSeek’s open-source large language models, which have a broad application potential, to improve work efficiency. So far, though, the AI assistant is not being used in core financial services such as bank account transactions, industry insiders said.

In the banking sector, chatbots are mostly used to enhance employee efficiency in areas such as code writing, internal AI office tasks and smart customer services, said Gao Feng, chief information officer of the China Banking Association. They are rarely used in core services such as bank transfers and other bank account transactions.

Bank of Jiangsu has successfully deployed and fine-tuned the DeepSeek-VL2 multimodal model and the lightweight DeepSeek-R1 inference model for smart contract quality inspection and automated valuation reconciliation, it said.

DeepSeek-R1 enables full-chain automation in the classification of emails, product matching, entry of transactions as well as the reconciliation of valuation statements, according to data from the Nanjing-based bank. The recognition success rate is over 90 percent.

It is still difficult to control the output of LLMs, an industry expert said. This makes it hard for them to be applied in the financial industry, and particularly the banking sector, which have high requirements for security and compliance. At present, AI assistants are mainly being used to generate content, which is then manually reviewed.

More than 10 public fund managers such as China Universal Asset Management, Fullgoal Fund and Lion Fund have said they are deploying the DeepSeek AI assistant in their systems. And Guotai Junan Securities, Sinolink Securities, GF Securities as well as a number of other brokerages have also said that they are applying DeepSeek-R1 in a number of scenarios.

The deployment of large models may become a common choice for financial companies, said Li Bolun, an analyst at Hong Kong-based Guotai Junan.

The financial industry has higher requirements for data security than other industries, and companies generally choose to store data locally, Li said. DeepSeek-R1 allows financial firms to combine data with large models at a relatively low cost to create enterprise-specific models.

In the insurance sector, China Ping An Insurance told Yicai it is actively researching and deploying the deep integration of big data open source platforms.

Editor: Kim Taylor