As discussed in previous Lanware articles on artificial intelligence (AI), there has been a huge explosion in large language models (LLMs) built for multi-purpose questioning and content generation. These systems are both powerful and popular and are being used for chat purposes, content and code generation and in other areas including the development of ideas in marketing and PR.
When looking at utilising these systems for work purposes, it is important to ensure you understand the end user terms of service and most notably understand clearly that any data you put in, can, and most likely will be used as further training for the system. Do you have a clear understanding as to what might be confidential when using these systems?
Using Microsoft Azure to create your own private ChatGPT
One way to ensure you can both utilise the power of these systems whilst keeping company information confidential is to build one yourself. As one of the core trailblazers in this area, ChatGPT can now be built in isolation using the Microsoft Azure platform and by doing this you can take sole ownership of the questions you ask.
In addition, a particular advantage is to manage how the system learns and controls the data it has access to. All LLMs are “dumb” when first deployed. By “dumb” we mean simply that they have not been given any data to train on or use for interactions with users. When understanding that most publicly available services are generally trained on a broad range of open data (usually via the World Wide Web), it is easy to see that restricting this system data could be hugely powerful for your organisation.
As a Microsoft Cloud Partner, Lanware has already started building test LLMs using Chat GPT in Azure and early outputs are showing huge potential. By deploying your own private LLM in your financial services business, it is possible to choose exactly what you train the service on and the data it has access to.
How could this be used in the financial services industry?
Following are four example use cases for private LLMs for financial services:
- Customer support: LLMs can converse with users accessing account or investment information. As a text-based questioning system, this can replace a user at the other end of the keyboard. As with any live platform, security and user authentication processes need to be robust to ensure customer data stays private.
- Product and service recommendations: Investment managers go through a careful process of assessing a customer’s investment needs and risk appetite. Teaching a system the processes involved in making recommendations could facilitate at least some of this process.
- Data interpretation: ChatGPT could be used to quickly interpret complex financials using earning reports, balance sheets and looking at overall sector market trends to help assess potential investment opportunities.
- Portfolio recommendations and ratification: An LLM could assist a portfolio manager in making investment decisions or ratifying ones already made. Specifics around the ratification process could be controlled by the data used to train the system.
As with all decisions around your future use of AI, we recommend a cautionary approach and ensure that any system considered complies with all regulatory, ethical and policies appropriate to your business. We wrote recently in detail about this area here.
