Episode Details

Hope you had a wonderful start to the new year.

This episode is special for multiple reasons.

This is one of the few episodes when I had the conversation with a guest in person - and that was also video recorded, with a fireplace in the background, on a cold day!

You would be able to watch this conversation as a video on the PM Power youtube channel.

This episode was recorded in November, 2022 - much before ChatGPT was made available to the public and the subsequent increased interest.

My guest today, Steven Aberle, is a founder of a company that is in stealth mode, working on domain aware, Large Language model based processing for very niche applications.

In this conversation, Steven shares

• How he joined the military half way through college and then went back to finish his studies
• Getting interested in IT during the military service
• Navigating away from the back end to the front and, focusing on how the users interact with the systems
• Being in the Washington DC area, getting to work on Defense related areas
• Working on defense proposals, that need to be technical, unique and bring out the uniqueness of the proposed approach
• Creating a piece of patented technology for domain aware generative AI, based on Large Language models and transformers
• The challenge for small technology companies to break into the defense procurement space
• Couple of challenges with the open LLMs
• Trained on open content available, thereby missing contextual insights that are more valuable
• None of the data that is generated, can be attributed to the source material [explainable AI]
• How he organizes his own learning even when moving across specialization areas
• It is always about the data
• His exposure and experience with ETL approaches, including working with unstructured data
• About being a full stack engineer, to be aware of many areas
• Not be as much of an expert as one may want to be in any one area, but connect the learnings to the new areas as well
• His approach for being a T-shaped individual, going deep into LLM, on how they are trained 
• Transformers are the next big wave now, after the cloud a couple of decades ago
• The concept or role of a solutions architect and how that may be implemented in a large language model