A Week in the life of AI

February 4, 2025 | Tom Jordan, President of J&J

Just as I was contemplating whether OpenAI, Meta, Google, Microsoft, Anthropic or Elon Musk (yes, that Elon Musk) was leading the charge in AI, you learn about DeepSeek, a made in China AI model.

DeepSeek reported that training one of their latest models costs $5.6 million, compared to the $100 million to $1 billion range previously estimated by US companies.

The Artificial General Intelligence (AGI) business evaluation is so significant that the market can  lose $1 trillion in one day when competitive alternatives become known. It is also critical and enjoyable to follow for business, military and generational leadership reasons. But then I must go to work.

Last week I interacted with the practical usage of AI, that brings value to our capital markets.

My first encounter was with a project team analyzing market data license agreements, to help clients efficiently understand how they can utilize data within their organization.

They were training a large language model to understand the nuances of terms such as “derived” which can also be called “new original works” or “transformed data”, and also to understand where such data can be distributed, whether to their own employees or to other clients in data particular regions, according to the license.

 This application uses two models, one for categorization (grouping data into predefined categories) and the other for inference (drawing conclusions). Both have been adapted from open-source base models, chosen for their unique strengths, and further trained using proprietary data.  The team has not evaluated DeepSeek yet because they are so close to a deliverable and that would distract them. A common thread between AI and traditional development, is that you have to ship to get revenue.

The second AI encounter was with our Regulatory & Compliance lead who is building a surveillance service with a partner company.  The goal in this instance is for AI to identify anomalies in data or patterns not detected by specific scenarios.  Additionally, using AI in analyzing alert data will be extremely helpful in identifying bad actors.

The last encounter is not in an advanced stage.   At a Research Consortium meeting, member firms were discussing how best to identify the parts of research content that may be subject to different rules regarding usage in artificial intelligence tools. As sell-side publishers begin allowing trusted vendors and buy-side partners to feed their research into artificial intelligence tools, there may be some content in the research – such as 3rd party data – that will need to be excluded.   Sell side publishers, buy-side consumers, and vendors providing content creation and aggregation services  were all part of the conversation, and the collaboration regarding both the business and technology considerations will continue.  

The bigger picture is more fun to read about. But our jobs are to use innovation that assists our business revenue growth and focuses on expense reduction.  If built and delivered effectively, they can provide significant benefits in 2025.