Sign In

Nick Elprin

Nick Elprin is a seasoned investor and advisor with a focus on Fin Tech and Media companies. He brings valuable experience as the former Head of Research at Bridgewater Associates to his current roles.

Having pursued both a Master of Science degree and a Bachelor of Arts degree in Computer Science from Harvard University, Nick Elprin boasts a strong educational background in technology and analytics.

Currently, Nick serves as the Co-Founder and CEO of Domino Data Lab, a position that allows him to leverage his expertise and entrepreneurial spirit. He also held the role of Senior Application Technologist at Bridgewater Associates in the past.

Highlights

Mar 28 · linkedin

Last week, NVIDIA GTC took over the San Jose Convention Center, the same location that hosted Strata (the "big data" conference) a decade ago. Walking through the expo hall, I was reminded of the opening sequence in the show "Silicon Valley": a time lapse of companies emerging, disappearing, or changing their logos. Back during the Strata days, every company was a "big data" company. Today, every company is an "AI" company. Standing in the expo hall, I imagined fast-forwarding over the last decade while booths replaced "data" with "AI" in their signage.

Of course, the potential of AI has already proved more real than the hype around big data. But in the midst of all the AI hype, there are some lessons we should remember from the "big data" era, to stay grounded at a time when it’s so easy to be swept up by the latest trend.

  1. Technology is always a means, not an end. Leaders who have the most success with AI will be the ones who start from the business outcomes they are trying to deliver, and work backwards from there. Just like the collecting tons of data didn't magically transform businesses, sprinkling LLMs everywhere won't realize the full potential of AI.

  2. In a large organization, people and process are bigger leverage points than incremental technology improvements. The highest ROI investments will be things like training employees; ensuring teams have the right skillsets; equipping people with productivity-enhancing tools; defining standard ways of working; increasing collaboration and knowledge sharing to maximize shared context and reuse of work.

  3. It’s easy to make a mess that’s hard to undo. When a new capability causes a frantic flurry of activity, companies create organizational debt that becomes hard to unwind later. Duplicative work, processes with opaque dependencies, and tribal knowledge were all problems that arose as companies blindly collected data — and they will likely arise as companies spread AI everywhere.

  4. From a safety perspective, less is more. Data and models can be assets — but also liabilities that create risk. It’s critical to create controls to ensure these assets are created and used appropriately — without being so onerous that you stifle innovation.

“Feeds and speeds” — the technical stats, e.g., model size — are easy to focus on because they are so concrete. But they are far less important than the factors above.

Mar 22 · linkedin

It’s easy to get caught up in the hype of model size and chip performance as a way to accelerate the impact of AI. These stats are the new “feeds and speeds” of the AI era. But in any enterprise with a large data science team and a wide range of AI use cases, optimizing your organization’s people and processes will be a bigger leverage point.

To that end, here is a simple framework and approach to help companies deliver more impact from AI.

Think about your DS & AI organization as a factory with two functions: Prioritization and Delivery.

“Prioritization” involves selecting a portfolio of projects with a good mix of risk/reward profiles. There are likely many “POCs” for use cases that have to be prioritized and narrowed down to turn into real projects.

“Delivery” is about executing on the projects you’ve prioritized, as quickly as possible.

Your aggregate business impact is the impact of all the projects you deliver in a period of time. So there are four factors that determine that business impact:

  1. The quality and quantity of your POCs (Prioritization)
  2. The quality of your prioritization (Prioritization)
  3. The time it takes (“latency”) to complete a project all the way to business impact (Delivery)
  4. The number of projects (“bandwidth”) you can do in parallel (Delivery)

By asking, “what would improve performance” of each of these factors independently, AI leaders can build a roadmap to significantly increase the impact of their AI organizations.

For example, to reduce average project latency, chances are that streamlining data access and discovery will have a bigger impact than using a more powerful GPU. Similarly, enabling more accessible and reliable elastic compute capacity may increase the team’s ability to do more in parallel, thereby increasing bandwidth.

Nick Elprin
Add to my network

Location

San Francisco, California, United States