Open AI
Google Brain
DeepMind
Meta AI
Stable diffusion
Multilingual machine translation
GPT3
The natural next thought is around chat bots / customer service / customer experience.
However, we’ve seen some pretty bad applications where chat bots don’t fully understand subtle inflections and latent context, which is kind of the whole point of that department in a company. Everyone gets mad, from customer to agent.
The fundamental building block of any form of generative AI is data. But not only is ChatGPT not trained on financial data (rather it is an expert on everything on the internet and programming), data in finance is very unlike data in computer vision or NLP. Financial data is extremely rich, but organized in an inconsistent way, with inconsistent rules, to encode lots of implicit information. Basically, the data structure is not easy to navigate, and financial data doesn’t have the absolute nature that physics does because all the rules around it were created by us wonderful humans.
So is it even worth it with such large potential downside? Well, yes. We aren’t there right now but there is (obviously) huge opportunity in better data & analytics for financial institutions, from fintechs to insurtechs, the latter of which is completely dependent on risk modeling.
Some successful surface level opportunities already exist, primarily in the lead gen space and sentiment analytics e.g. Gong, Observe, etc as well as algorithmic trading e.g. Renaissance Technologies and asset management portfolio construction. And perhaps another lower hanging fruit is generating insights from transaction data, as it can help humans do their job better.