Problem-Solving
Writing down this thought about problem-solving.
I think that as our society progresses, and especially in the field of Engineering, it is much less about your ability to memorize things, and more of your ability to learn fast and understand how you can use different concepts to solve a particular problem.
This is why I am grinding out Competitive Programming.
YouTube is like the ultimate way of Problem-Solving, because I get to choose problems that have not been solved before, and it’s just more rewarding.
Younger people rely on fluid intelligence (capacity to adapt to new situations and use trial and error to quickly figure out solutions)
By contrast, older people tend to rely on crystallized intelligence (use superior store of knowledge to solve problems).
Resources for better Problem-Solving
- Mental Models
- I feel like these mental models really help solve more global social problems than they do local technical problems.
Specific Problem-Solving Scenarios
This is extremely annoying.
When you change something to fix something else, but once that thing is fixed, something else is broken.
Examples:
- Overfitting
- Tuning Tesla Autopilot, neural net does better for traffic sign detection, but now is really horrible at detecting pedestrians
LLM Problem Solving
When I solve problems are work, what I am trying to do is essentially gather insight and then incrementally solve the problem. I am reasoning through the problem and figuring out what knowledge I require to solve the problem.
In engineering, there is always uncertainty. You should know how to deal with those uncertainties.
LLMs lack that capability.