Qirui > Recent Status Updates

1. Optimisation is not optimal if the problem is computationally complex or search space chaotic.
2. Heuristics aren’t limited, they are fast and robust.
3. Heuristics can be used for big, small, important, unimportant problems.
4. Less information and/or computation can be better - reduces overfitting, faster answer.

Most drivers are safer on average, accidents disproportionately occur to very bad drivers!

1. Optimisation is not optimal if the problem is computationally complex or search space chaotic.
2. Heuristics aren’t limited, they are fast and robust.
3. Heuristics can be used for big, small, important, unimportant problems.
4. Less information and/or computation can be better - reduces overfitting, faster answer.

Most drivers are safer on average, accidents disproportionately occur to very bad drivers!

Light, amusing and gets me thinking.

Chapter 9 Inference Optimisation is especially focused on LLMs and not entirely relevant for image models.

Lighthearted and thought provoking.