The "buy vs. build" debate is evolving into a more nuanced discussion about control, scale, and security, especially when it comes to models. At Capital One, our approach to AI innovation has anchored around deeply customizing open weights models with proprietary data. Customized models deliver better compute efficiency, latency, and performance, resulting in better experiences for associates and customers while ensuring a balanced, well-managed approach. This has manifested into some transformative outcomes, including the launch of our first customer-facing agentic use case last year. I enjoyed sharing some perspectives on this topic with Nat Rubio-Licht and The Deep View recently. Check out the full article here: https://lnkd.in/gGZUFD5g Prem Natarajan, PhD Robert Alexander
Oded Cohn
1w
Good to hear from u Milind; makes perfect sense
Interesting read. One question it raised for me: once you heavily customize an open model, the harder problem may become ongoing assurance and governance after modification, not just model selection itself.