Lately, while diving into the latest and newest techniques in the world of AI, I couldn’t help but notice a few concepts that bear striking similarities to life lessons we learn as humans.
-
Reinforced Learning with Human Feedback (a.k.a. RLHF): In AI, this technique fine-tunes LLM’s performance based on human feedback. In real life, as we grow through school and society, our belief system is constantly fine-tuned. The people around us shape our worldviews.
-
Prompt Engineering & Soft Prompting: While these are distinct techniques to help LLM provide relevant information, it is akin to communication skills we learn. Even if you’re knowledgeable, it’s crucial how you share that knowledge. No matter how smart or informed you are, if you can’t convey it effectively, it loses its value.
-
General vs. Domain-Specific Training: Should a model be trained broadly or specialize in a particular area? Same question can be asked when we pursue an education or career. The answer is nuanced: to truly excel in a specific domain, you need a broad base. For instance, in order to get more intuition about advanced ML techniques, it helps to know how derivatives and probability work.
-
Balancing Parameters, Training Data Size, and Total Compute Cost for LLM performance: While an AI model is not bounded by how long they live, we are. Consider our limited time on Earth. How do we make the most of it for our friends, family, and society?
-
ReAct & Train of Thoughts: Knowledge alone is not enough, same for LLM and same for us. Probably a realization felt by many folks who endured the journal of obtaining a PhD. Often time, the most valuable thing you gain from the degree is not the science and knowledge itself, rather it is how you think and approach problems.
This post was originally published on LinkedIn1.
If machine learning mirrors life lessons, you might enjoy The Long Run about finding meaning through persistence and iteration.
References
Footnotes
-
Wang, L. (2024). Life Lessons from Machine Learning. LinkedIn. ↩