Suggestions
Andrew Barkett
Founder, RAG AI startup, Former FB+GOOG. Investor/advisor to startups
Andrew Barkett is a seasoned professional with a robust background in General Management and Technical Architectures. He has excelled in various settings, spanning from startups to established market leaders. Andrew's expertise lies in software architecture, leading technical teams, building distributed systems, managing hyperscale datacenters, cultivating team growth, and mentoring startup founders. He pursued a Bachelor of Arts in Political Economy at the University of California, Berkeley, and later completed an MBA with a focus on General Management at the University of California, Davis - Graduate School of Management.
Highlights
AI thought of the day:
“There is another world and it is this one.” Probably from the French Surrealist and Communist poet, Paul Éluard.
"In essence, LLMs reveal a multifaceted and layered world that exists within our current reality, much like the quote suggests. They bring to light new ways of thinking, interacting, and creating, transforming our perception of what is possible and redefining the boundaries of our world. Through their capabilities, we access a richer, more complex dimension of existence that was always there, waiting to be uncovered."
The above explanation came from the ChatGPT, using this prompt: “There is another world and it is this one.” Probably from the French Surrealist and Communist poet, Paul Éluard. Relate that quote to the role of LLMs in modern society"
Here is the key LLM insight you need today.
What percentage of human communication is genuinely novel, creative, or truly analytical, as opposed to simply predicting the next words that seem appropriate in a situation?
Example 1: "Hi, how are you?" "Fine, thanks, just walking the dog." --Is that really "how you are?" Does it contain anything except a boilerplate script you're following, with a little bit of context?
Example 2: "What should we put on slide 8, titled, "marketing plan?" "Let's use some industry standard rates for these assumptions and build a simple table that shows marketing spend by quarter and projected sales." "No, those numbers seem unbelievable, make them a little lower." --Is that really different from what the LLM is doing when you ask it to create a marketing plan?
The mistake people make isn't that they're overestimating the ability of AI. It's that they're overestimating the substance of most human communication. Most of the time humans are doing nothing more than predicting, "What words would sound good next?"
An exception to this idea that the LLM actually does approximate what a human would do pretty well comes when you need the AI to act on 'private' or new data. That's where RAG comes in. That's why it's the most important direction right now (more important than fine-tuning, more important than agents). Assume the foundation models will get better.
