In an era where artificial intelligence is reshaping industries, the financial sector stands at the forefront of this transformative wave. At the heart of this evolution is Aleksandr Ogaltsov, a Data Scientist at Kodex AI, who graciously shares insights into the dynamic synergy between cutting-edge AI technologies and the intricate landscape of finance.
Through a candid discussion, Aleksandr unveils the pivotal role of data science in shaping tailored solutions for the finance industry. From fine-tuning Large Language Models (LLMs) to the ethical considerations entwined with AI utilization in finance, his perspective delves deep into the nuances of leveraging technology while preserving transparency and trust.
Join us as we navigate through the realms of AI's impact on financial operations and explore the potential it holds for empowering financial professionals.
What role do you play as a Data Scientist at Kodex AI?
As a data scientist, I’m constantly working on keeping up with the advances of science and technology behind our products. The other half is to creatively combine and tailor those advances to the very particular needs of our customers. Finally, with support from the rest of the development team, we deliver to a challenging production environment – mainly ensuring as low latencies as possible while maintaining high standards of data security which is crucial for our customers.
Why is there a need to fine-tune LLM, especially for the finance industry?
LLM stands for Large Language Model, which literally means that we are building a computer model of some language(s). Although language models have existed for a long time, a real revolution happened when LMs became truly large – people discovered that, as a byproduct of the size, there are super useful capabilities like accomplishing natural language tasks in a zero-shot fashion (which is broadly called prompting). However, the language for which we are building a model matters. Even from an everyday life experience, we can feel that the language of many domains of human knowledge varies a lot. For human experts, it usually takes time to onboard themselves simply to the language even of the quite close domain. The same roughly holds in the case of building a computer expert – we have to tailor it to a financial industry’s language. In the case of the financial industry, it’s actually not only special terms, numbers, and their interrelations, but the layout of the information: from complex tables and figures to spreadsheets with financial data – all of this makes it crucial to build for the industry individually, fine-tune and specialize.
Are there some potential ethical concerns and challenges associated with using AI in finance?
The major challenge is to build trust between customers and the product. Since we build a decision-making support system for financial experts, it should be clear and visible based on which information decisions are suggested and which exact intermediate steps are taken to make a certain decision recommendation. Because in the end, it is the expert who is responsible for the final decision roughly with their signature, not a computer model. However, a computer model can be transparent enough to support and truly accelerate complex decision-making processes. That transparency level can be achieved by fine-tuning to get rid of so-called hallucinations as well as insertion of the transparency and visibility mechanisms into the product.
Can you provide a few examples of AI use cases that financial professionals can benefit from?
I’d say document drafting is the most promising use case. Repetitive, but at the same time always somewhat different customer questionnaires, requests for proposals, etc. In my opinion, this is an exact hotspot for the recent AI advances – too much complex task for the previous generation of natural language models, but at the same time one of the most time-consuming chore tasks for financial people. This combination makes it a good entry point to bring perceivable value for the customers.
Finally, how do you stay up-to-date with the latest advancements in AI?
Arxiv papers digest + following papers from top-tier ML conferences + making HuggingFace my default social network to scroll 🙂
.png)
