What’s the next step of AI utilisation?
When end users like traders need to make split second decisions for a living, they want to take advantage of all the information that’s available. Natural language processing is a technology that has been in existence for several decades, however, the issue has been the slow processing times – it wasn’t fast enough to be useful. The technology has gone through a transformative shift, making it possible to answer complex questions very quickly from vast data sources. However, the challenge remains around the seamless blending of disparate data sets across different systems to arrive at an answer.
I think that the possibilities are just beginning to be understood. If you look at the pace of change of a company like OpenAI, they had been around for 12 years when they launched ChatGPT, and then everyone learned their name as it became the most used application in the history of software. There are a lot of people experimenting.
Will there come a time where AI plays more of a role in how to execute the trading decisions?
The potential for AI tools to play a role in trade execution remains unknown. When considering AI’s role in shaping trading decisions, you can think about a period several years ago when algorithms were newly introduced to the equities market. At their inception, they were relatively rudimentary, consisting of just two fundamental algorithms. Over time, major dealers introduced more algorithms, which led to a common question: “What is the optimal algorithm for a given trading scenario?” It’s a common question in electronic trading across most markets, where a wide array of tools is made available. The question remains: “What is the most suitable tool for a specific task?” and AI may help to answer that question.
To date, what we are witnessing at LTX is individuals utilising our GPT-powered application BondGPT to ask questions related to liquidity availability, counterparty selection and bond selection. Subsequently, they employ our RFQ+ and RFX trading protocols to assist them in the execution phase. Will individuals leverage GPT systems to facilitate their trading activities? It’s a question that we are exploring with our clients and our advisory group.
What were the empirical hurdles for incorporating AI (GPT)?
One of the most common concerns about GPT technology was hallucinations. GPT, by design, strives to be accommodating, which doesn’t suit financial market participants who require accurate and verifiable information. In the case of BondGPT, we looked to the success of ChatGPT in terms of a simple, easy to use natural language interface but focused on curating the data very carefully. To meet the needs of financial markets users, we need to ensure that only the highest quality sources of data go into providing answers and that there is no creativity coming from the generative aspect of GPT and creating hallucinations.
We had another matter that we also needed to resolve which was compliance. We operate as a broker dealer, like most electronic trading platforms, so we needed not just to source an answer from the data, but also be careful not to violate any compliance rules in addressing the questions a user might ask. So not only did we need to curate the right data and make sure that we could deliver the accurate answers quickly, but we also had to make sure that the answers were compliant.
In developing BondGPT, we also quickly realised that we needed to incorporate training on the end users’ vernacular to deliver a usable product. In our market, traders expect to be able to speak the minimum required to be understood, so we ensured that the model understands bond market jargon common to trading desks. Another issue with GPT was that it didn’t do math. In fixed income, bond math is just part of answering questions that users need answered to help do their jobs, so we needed to combine high quality data with the ability to do accurate calculations.
So again, we needed to take the essence of the way ChatGPT was serving users and what made it so popular, and we needed to apply it to financial market participants and fixed income in particular.
What do you think of the notion that the AI bubble might burst soon?
I have a different perspective. Conversations around the applications of AI have taken place for many years now and many believe there is still a lot of potential to be had. However, its practical implementation was hindered by the limitations of computational power.
In fact, AI has actually been incorporated into products that people don’t realise, but now it’s sparking imaginations because it can now do some things that are really difference-making in people’s lives and roles. People are just now discovering some of the new uses and I think that, if anything, the pace of change and the incorporation of AI is going to accelerate from here.