ML systems debugging rewards a willingness to dig across the entire stack, to chase a slightly suspicious signal back to its source, and to derive chains of failures from surprising end results. High cognitive burden but also some of the most exhilarating work upon success.

May 27, 2022 · 4:06 PM UTC

12
26
5
279
Replying to @gdb
Your Twitter handle would suggest you enjoy debugging, so checks out
Replying to @gdb
They are geniuses, blessings more wisdom may the Lord give you
Replying to @gdb
Couldn’t agree more.The only difference from software 1.0 style debugging is the learned mind shift that many a times the cause could be a hyper parameter tweak, or “improperly cleaned data” more than an actual code change,while the latter also happens even if only infrequently.
1
Replying to @gdb
Substack is sometimes more challenging than stack.
Replying to @gdb
specifics please..
Replying to @gdb
I miss chasing a crash due to a corrupted pointer due to negative index access on another array on an embedded system with simple RTOS where you only have terminal access through a UART in 90's.
1
Replying to @gdb
Agreed 🥂
Replying to @gdb
Brick by brick
Replying to @gdb
"ML systems debugging rewards a willingness to dig across the entire stack, to chase a slightly suspicious signal back to its source, and to derive chains of failures from surprising end results" 1 "High cognitive burden but also some of the most exhilarating work upon success" 2