3 ms·

Harness, or train? When a junior uses AI the junior learns. When a senior uses an AI the AI learns. Maybe the AI owners are harnessing experienced devs to train

Harness, or train? When a junior uses AI the junior learns. When a senior uses an AI the AI learns. Maybe the AI owners are harnessing experienced devs to train their replacements.

Rover222 1y ago
That kinda made sense except for "When a junior uses AI the junior learns"
jdlshore 1y ago
“AI”—assuming you mean LLM-based coding assistants—don’t learn. All they have is the context you give them. Seniors may learn how to manipulate that context to get better results over time, but that’s not the AI learning in any meaningful sense.
- 1W6MIC49CYX9GAP 1y ago
  The LLM answers your accept/reject are used as training data
  - jemmyw 1y ago
    That usually isn't true if you're using the agent tools aimed at companies. They are explicit about what is and isn't used for training.
godelski 1y ago
I think this generalizes poorly. A junior can use an AI to learn (just like anyone) but a junior is also much more likely to feel the pressures of getting things done quickly and not taking the time to understand what was written. A junior is also less likely to be about to identify when the AI makes errors as well as when accurately describe the demand to the machine. I see people refer to AI as "training wheels" but even training wheels force you to learn balance while riding a bike. They just give you more flexibility in how much you can lean. But AI, it is easy to just accept the answers and treat it like a black box. There is no need to practice the actual skills of a developer: designing algorithms, analyzing, testing (designing tests!), reading code deep with a code base and learning how it connects, interpreting asks and determining appropriate ways to solve those problems (a big part of experience is having fewer unknown unknowns. Known unknowns are incredibly valuable) While all that is possible with AI, I don't see it happen often in practice due to other pressures. And with AI it seems there's just become increasing pressures to get more done faster. But a junior still needs time to learn. We all do, but juniors need more and have less experience and ability to find this time

Harness, or train? When a junior uses AI the junior learns. When a senior uses an AI the AI learns. Maybe the AI owners are harnessing experienced devs to train their replacements.

Rover222 1y ago

That kinda made sense except for "When a junior uses AI the junior learns"

jdlshore 1y ago

“AI”—assuming you mean LLM-based coding assistants—don’t learn. All they have is the context you give them. Seniors may learn how to manipulate that context to get better results over time, but that’s not the AI learning in any meaningful sense.

1W6MIC49CYX9GAP 1y ago
The LLM answers your accept/reject are used as training data
- jemmyw 1y ago
  That usually isn't true if you're using the agent tools aimed at companies. They are explicit about what is and isn't used for training.

godelski 1y ago

I think this generalizes poorly. A junior can use an AI to learn (just like anyone) but a junior is also much more likely to feel the pressures of getting things done quickly and not taking the time to understand what was written. A junior is also less likely to be about to identify when the AI makes errors as well as when accurately describe the demand to the machine. I see people refer to AI as "training wheels" but even training wheels force you to learn balance while riding a bike. They just give you more flexibility in how much you can lean. But AI, it is easy to just accept the answers and treat it like a black box. There is no need to practice the actual skills of a developer: designing algorithms, analyzing, testing (designing tests!), reading code deep with a code base and learning how it connects, interpreting asks and determining appropriate ways to solve those problems (a big part of experience is having fewer unknown unknowns. Known unknowns are incredibly valuable) While all that is possible with AI, I don't see it happen often in practice due to other pressures. And with AI it seems there's just become increasing pressures to get more done faster. But a junior still needs time to learn. We all do, but juniors need more and have less experience and ability to find this time