OpenAI, BloombergGPT, and Coca-Cola

As I wrapped up my book 2.5 years ago, I identified some opportunities in consumer credit risk management.

While I didn’t know about the work being done by OpenAI at that time, I narrowly focused on risk modeling and risk strategy.

I envisioned an AI bot with specialized risk domain knowledge, which makes risk management exercises faster and easier.

📌 Two years later, 𝐎𝐩𝐞𝐧𝐀𝐈 released ChatGPT.

Just in time for a world that badly needs more productivity to tame high inflation.

It came in a different form from what I envisioned - a fundamental large language model (LLM).

It still amazes me how it can present knowledge simply by focusing on predicting the next word.

ChatGPT being a generalist is already impressive. How about instilling the subject matter knowledge?

📌 Here comes the 𝐁𝐥𝐨𝐨𝐦𝐛𝐞𝐫𝐠𝐆𝐏𝐓, built on billions of token datasets from the financial documents by Bloomberg over 40 years.

It enhances finance-specific tasks, such as sentiment analysis and news classification, and unlocks new opportunities from the vast quantities of data available on the Bloomberg Terminal.

This is an example of domain-specific AI. I can imagine every consulting company, or company that owns a large amount of data, are busy building its own AI bots.

Anyone is building the credit risk AI?

📌 Recently Chamath Palihapitiya made an interesting analogy of OpenAI vs. domain-specific AI companies.

After the refrigerator was invented, the company benefiting most from the invention is not the refrigerator manufacturer.

It is the Buffet’s beloved 𝐂𝐨𝐜𝐚-𝐂𝐨𝐥𝐚.

Who is the new Coca-Cola in the AI era?

Which product is the new Coke that would occupy the refrigerator shelves?

I am glad to be at the moment - to witness the continuous advancement of human power and the shift of the business landscape.