Explaining how things work to different audiences is a topic that interests me. Can you put yourself in the mind-space of the person you’re speaking with — and explain a concept — so that they truly receive what you’re saying? I’m also curious about the view that true value in LLMs is through getting people to use them in the problems they encounter daily — ‘AI meets life stuff’.

LLM infrastructure value will be captured by the major players (Google, Meta, Microsoft, etc.) — but the application play could have a lot of winners. How do we get people — who do work that could benefit from LLMs — aware enough to explore them and find the right usage fit?

One way is to help people understand that LLMs aren’t magic. When something seems like magic, people often find it hard to reason about how it fits in, or relates to other things. Magic can paralyze your creativity / relation-making abilities. Life is pattern matching - if you don’t understand or see the pattern, it’s hard to match it to something else. When something is magic, you assume what you’re seeing doesn’t relate to anything else you know about. Case in point with a personal story - leading up to my open heart surgery in 2019, heart lung bypass seemed like magic, and thus felt more terrifying to me. How could this possibly work? I didn’t even want to think about it. Thinking about it gave me anxiety. Learning more about how it worked gave me comfort. It made me more curious about other aspects of the surgery, and more confident I’d come out the other side ok. Removing magic gave me power and clarity of thought.

Magic doesn’t really exist - all magic is just a comprehension gap in the receiver’s experience. Effectively explaining how something works , in language that the audience understands and relates to, can scrape the magic off. It can make it easier to reason about, think about, and take action on. How might we do this for LLMs? I’ve been riffing on the following hypothetical as a way to describe how LLMs work:

  • Imagine you ask a question out loud, and everyone in the world can hear your question and begins responding.
  • Everyone starts answering the question, but they each say one word at a time.
  • You are able to hear everyone’s first word, and can understand what the most popular word was
  • This word becomes the first word in the answer
  • Then, everyone is informed of the first word in the answer, and is asked to say what the second word is. The most popular word is again chosen
  • The process repeats until an answer is generated

I don’t think the above is perfect, but it might get closer to scraping off the magic, and helping more people understand how they might use LLM capabilities in their daily tasks.