In the whole deluge of new generative AI models like ChatGPT being released by tech companies, one significant announcement was missed by many: the release of BloombergGPT by Bloomberg. The leading purveyor of financial data built a fifty billion parameter Large Language Model (LLM) using their huge datasets to finetune the LLM from scratch.
As a comparison GPT3 has 175 billion parameters, but Bloomberg’s data was purely financial. Combine this powerful financial-focused foundational model, with the amazing conversational abilities of ChatGPT, and we could pretty much have answers to any financial question. So, can Generative AI give you financial advice, can you rely on it, and could it potentially replace human financial advisors?
Today, if you go and ask ChatGPT Plus (which is GPT4) on whether I should buy Apple stock or not, it flat out refuses to do so and asks you to consult an advisor. If I make the question more generic and ask it how the stock will move, again it demurs.
There are two reasons for this: One, ChatGPT has built guardrails around ChatGPT to not answer these questions both for ethical and legal liability reasons, and two, it is impossible for ChatGPT to do so, since it is not connected to the treasure trove of the World Wide Web. To control ChatGPT, OpenAI stopped feeding it data after 2021.
However, a specialised product like BloombergGPT and the many others which will come out might not have these limitations. Having said that, I do not see Generative AI, at least in its current form, offering reliable financial advice.
AI has been used extensively in the financial services industry, perhaps more than any other, to predict prices, shifts in market, etc. and has been offering ‘financial advice’ to professionals for many years. We have been doing algorithmic trading, where AI based algorithms trade by themselves, without any human agency.
However, the new wave of AI, which is Generative AI, has some problems. It is built differently, as large auto-complete language engines rather than pure machine learning based enterprise AI. Generative AI is optimised for plausibility not truth, it often hallucinates, and makes stuff up as it goes along. This is certainly not conducive to offering sound financial advice. As models do get better, and specialised models like Bloomberg’s come up, however, we can expect some reliability improvements.
I have always maintained that AI will not take our jobs, but human beings using AI could. Thus, a financial advisor who uses these powerful tools, could offer better advice to you than someone who does not. She could have AI bring out the right financial data, present it in a simple fashion, give insights which she might miss and present various advice options. Then with her experience, intuition and customer knowledge, the advisor could present the best advice much faster.
This is why a company like Morgan Stanley is experimenting with GPT4 at scale. It has identified content in over a hundred thousand internal documents on which its over financial advisors could have questions like investment recommendations, general business questions and process questions It is then fine tuning GPT4 to help answer this.
So, this is how I believe that reliable financial advice will be provided: by AI and humans working together.
-The author is founder, The Tech Whisperer Limited, UK. Views expressed are personal.
Read all the Latest Business News, Tax News and Stock Market Updates here
Comments
0 comment