Our Soar engineering team has been building custom ML models for years.
Since GPT-4 APIs came out, though, mostly we do prompt engineering instead of building custom ML models from scratch.
Today I got the first taste of our homegrown “Retrieval Augmented Generation” — a new subset of AI development — where you give AI access to a set of content in a vector database so it can do a semantic search (which is more powerful than keyword searching) and then GPT4 analyzes what is retrieved and throws out the stuff that doesn’t really match the intent of your search.
Experiencing this today — conversationally asking some questions about the people and topics discussed on more than 1200 recorded zoom calls (since late 2021) — is one of the most powerful computing experiences I’ve ever had.
I uncovered two huge leads from over six months ago that had fallen through the cracks. Each of them could be a six-figure deal. And the AI can point me to key things they said on the calls which can lead to the next step.
It’s truly a step towards Jarvis AI.
