**Hangzhou**: DeepSeek launches its multimodal large language models, DeepSeek-R1 and R1-Zero, boasting 671 billion parameters and aiming to outpace OpenAI. The move reflects China’s push for AI innovation amidst restrictions on Western models, reshaping the insurance landscape with open-source benefits.
DeepSeek, a prominent Chinese company founded by Liang Wenfeng in Hangzhou, Zhejiang, has recently made headlines with the unveiling of its multimodal large language models (LLMs) on January 20. The new models, named DeepSeek-R1 and DeepSeek-R1-Zero, are set to compete with OpenAI’s o1 model and boast an impressive open architecture with a total of 671 billion parameters. This launch follows the company’s earlier release of DeepSeek V3, which was more akin to OpenAI’s GPT-4. The evolution from V3 to the R1 model signifies a shift towards a reasoning-based model, highlighting the progress made by Chinese AI researchers.
The rapid advancement of AI technology in China can be attributed to at least two major geopolitical factors. The restrictions on access to Western AI models—as of January 2025, OpenAI’s and Meta’s Llama models remain inaccessible in China—have prompted local companies to foster their own innovations. Alongside this, hardware constraints due to U.S. export regulations limiting the availability of high-performance Nvidia Graphical Processing Units (GPUs) have necessitated a search for efficiency. Despite these challenges, DeepSeek has reportedly achieved success in training its V3 model with only a fraction (5%) of the GPUs that OpenAI utilised, suggesting a competitive edge in operational efficiency and cost management.
Moreover, the distinction between open and closed models is proving crucial for industries like insurance, where factors such as transparency, interpretability, and compliance are critical. Closed models—including OpenAI’s GPT models and Google’s Gemini—keep their proprietary data out of reach, only accessible through APIs. This has raised concerns among insurance companies regarding data privacy and compliance issues. Conversely, open-source models like DeepSeek’s offer enhanced transparency and control, as companies can deploy and customise these models on their own infrastructures.
In the realm of life insurance, the challenges posed by unstructured medical data have garnered attention. Insurers often grapple with underwriting processes that depend heavily on diverse forms of medical documentation. Fine-tuning domain-specific models presents a cost-effective alternative to relying on expensive, API-based closed models. Use cases for these open models include:
-
Medical Document Analysis: By fine-tuning an open model on historical underwriting data, insurers can significantly improve the accuracy of extracting crucial information from the broad array of unstructured medical records.
-
Fraud Detection: Training models on anonymised claims data can reveal patterns that may indicate fraudulent activity, ultimately helping to mitigate the costs associated with false claims.
-
Customer Service Automation: Customised chatbots can be deployed to assist policyholders with inquiries and claims processes, enhancing response times and customer satisfaction through targeted fine-tuning.
As insurance companies increasingly turn to generative AI models, data security remains paramount. Running an open-source model in-house or via a trusted cloud provider—such as AWS, GCP, or Azure—ensures comprehensive control over sensitive policyholder information. This contrasts sharply with using external APIs, which often involve contractual scrutiny concerning data usage.
In response to this growing landscape, best practices have emerged to ensure data privacy while adopting AI models:
- Deploy models internally or utilize a trusted cloud infrastructure to safeguard sensitive data.
- Carefully review legal agreements with API providers, ensuring stipulations prevent proprietary data from being used in model retraining.
- Establish strict access controls and audit processes for AI-generated outputs to maintain compliance.
The rise of open AI models positions life insurers to take advantage of transformative opportunities, from cost reductions to enhanced accuracy and data privacy. The advancements made by companies like DeepSeek suggest that the landscape of AI is leaning towards open innovation, allowing for a more competitive and equitable environment. As the insurance sector continues to adapt, forward-thinking companies are likely to capitalise on the potential that generative AI holds in driving operational efficiency and business growth.
Source: Noah Wire Services
