Can ChatGPT Actually Work for Investors?
Undoubtedly, most of you are aware of or already using generative AI through various applications such as ChatGPT, which have been dominating headlines and mushrooming in popularity.
Since Omega Point’s inception almost a decade ago, we’ve always been keen to uncover and test new technologies with potential to improve our customers' experience and make us more efficient and productive in building amazing product and service offerings for professional investors.
Since earlier this year, Omega Point has been working with the team at Blueskys AI, who combine the raw conversational abilities of large language models, like ChatGPT, with domain-specific data, to build and train a generative AI prototype tailored to the investment community.
Initial results have exceeded our expectations, and you can learn more details about our initiative in the latest issue of Regius Magazine titled “2023: The Year of Generative AI”. You can also access a direct link to our article here.
Summary - Omega Point's Gen AI Initiative
Blueskys AI built an augmented language model that utilized our archive of 250+ issues of our Factor Spotlight blog posts. We ran an experiment using the Blueskys AI application, to answer 68 common questions received from customers and other investment professionals. BlueSkys AI ran the questions through two models:
- An off-the-shelf model using OpenAI GPT 3.5. This is the same model used by ChatGPT.
- Omega Point augmented GPT 3.5 language model built by BlueSkys AI
We then had subject matter experts from Omega Point score the responses to each question as “acceptable” or “not acceptable” for an investment professional.
The off-the-shelf model generated acceptable answers for ⅔ of the questions.
The Omega Point trained model generated acceptable answers for 97% of the questions.
In one example, we were keen to know how it handled broader, industry-related questions such as: What are hedge baskets?
Off-the-shelf model: "Hedge baskets are baskets used to carry or store hedging plants."
Omega Point trained model:
"Hedge baskets are custom baskets of stocks used to hedge out a particular portfolio or unique or proprietary factors. They provide full control to better meet the investor's objectives."
The off-the-shelf model has since caught up with a more relevant, albeit less amusing, response, likely from feeding its 'short-term memory' through our testing. But you get the idea...
Given this encouraging start, our team and BlueSky AI have been training the model using various inputs besides our blogs. In the nearer term, we plan to deploy this technology internally to help improve productivity and efficiency.
On the not-so-distant horizon, we envision leveraging generative AI to allow our customers to answer portfolio-specific questions or to educate themselves on the various methodologies for the datasets offered through our Data Cloud.
Click on the video below to see the model in action using a recent question from one of our Factor Spotlight readers.
We're excited by the many possibilities that generative AI offers to us and our customers and the ability of Blueskys AI to generate results that are more nuanced, relevant, and familiar to the investment community.
We hope you found our experiences and early results helpful in understanding your potential use cases leveraging this powerful and versatile technology.
If you'd like to learn more about this initiative or Blueskys AI, please reach out to us or Josh Lester at Blueskys AI.
P.S. This Omega Point blog entry was written in 98% human 😀