Beyond Will Hunting: The Need for LQMs in the Age of LLMs
An LLM, Large Language Model, is like Will Hunting.
We’ve all watched Good Will Hunting and relived its scenes countless times, using its numerous quotes and incredible dialogue (brought to life by Robin Williams, Ben Affleck, and Matt Damon) to explain aspects of life. The last time I was in Boston, I went to the Public Garden just to see the bench where Robin Williams delivers a life-changing speech to Matt Damon. This speech shifts the entire dynamic of the movie, breaking down the tough boy's defenses and exposing his vulnerabilities. Here it is:
"So, if I asked you about art, you'd probably give me the skinny on every art book ever written. Michelangelo? You know a lot about him. Life's work, political aspirations, him and the Pope, sexual orientation, the whole works, right? But I bet you can't tell me what it smells like in the Sistine Chapel. You've never actually stood there and looked up at that beautiful ceiling. Seen that."
An LLM is like Will Hunting. This scene highlights the difference between theoretical knowledge and real-life experience.
An LLM is incredibly smart and can quote any matter, author, any time, anywhere, and use it to its benefit—from art to business, history, etc.—everything that is in the books/internet. It’s a word repeater. But it can’t re-create what’s not in the books, although it can fool you and make you think it can.
How much is in the books or on the internet? If we could measure all the knowledge and data in the universe, less than 1%? In chemistry alone, there are trillions of possible interactions between molecules in the periodic table, and we definitely don’t have all of those documented anywhere, since trial and error has been our primary approach so far. So how do we access the other 99%?
领英推荐
Large Quantitative Models (LQMs) are foundational models that, instead of being built using words and images, use the equations that govern our world—from biology and chemistry to physics and mathematics. Running those equations allows us to uncover that 99% in a simulated manner, without having to experience it and then write about it in a book to teach at Cambridge to the next generation of students or the genius janitor.
LLMs can tell you the steps to design a new rocket fuel engine, however they can’t actually design it for you. LQMs can.
LLMs can tell you the most suitable sequence of proteins for a specific virus, however they won’t be able to run the equations that detail how they will bind and react. LQMs can.
LLMs can recreate a perfect copy of the Sistine Chapel however they can’t design a new material or paint to protect the actual Sistine Chapel from time and erosion. LQMs can.
Your move. Are you going to stay in your childhood neighborhood, repeating the words you can find on the internet, or are you going to take the road and discover what’s out there—the unknown universe only visible by solving scientific formulas with LQMs?
By the way, here is the famous scene: https://www.youtube.com/watch?v=oRG2jlQWCsY
Entrepreneur, Consultant, Mentor
3 周Congratulations Fernando and keep on sharing.
My LQM is wicked smaht
AI Mad Science & Deepfake Red Teamer
3 周Prompts must direct LLMs to use active voice, otherwise they invariably generate text in passive voice. Long sentences full of passive voice, kneecapped thesis statements, and every paragraph ending with some form of “this (X) highlights (Y)” — that’s how to spot LLM generated text. My favorite is the kneecapped thesis statements. An LLM will say “this (Z) could have ramifications to (A)” even if it’s clearly true; e.g., “the next epidemic could cause widespread economic damage”. What epidemic has ever not caused widespread economic damage? LLMs are BS generators. They’re not TRYING to lie to you. They’re TRYING to tell you what they infer you want to hear. If they can tell the truth they will. But, the truth is irrelevant to an LLM. What else is the definition of BS?