I think this goes for dependencies in general when approaching it from a development perspective. I remember implementing blowfish in 1996 by typing out the spec and test vectors in code. The spec was my dependency. Later it became OpenSSL. Nowadays it's a ton of potential implementations, depending on what I'm building for.
Does it mean "I can do cryptography"? Or "I can get it done"? I think the latter; I'm still not a cryptographer.
Agree with you. It's more general than just AI tools. It jus seems to be harder to trust and build confidence in AI tools than standard libraries as output are not deterministic.
This is why, when doing a data pipeline, I do AI-last. All the deterministic stuff goes at the front.
Even in coding this works best for me: let the AI act as a pre-reviewer on hand-written code. Like an artificial second set of "eyes" before I bother any real eyes.
I think this goes for dependencies in general when approaching it from a development perspective. I remember implementing
blowfishin 1996 by typing out the spec and test vectors in code. The spec was my dependency. Later it becameOpenSSL. Nowadays it's a ton of potential implementations, depending on what I'm building for.Does it mean "I can do cryptography"? Or "I can get it done"? I think the latter; I'm still not a cryptographer.
Agree with you. It's more general than just AI tools. It jus seems to be harder to trust and build confidence in AI tools than standard libraries as output are not deterministic.
This is why, when doing a data pipeline, I do AI-last. All the deterministic stuff goes at the front.
Even in coding this works best for me: let the AI act as a pre-reviewer on hand-written code. Like an artificial second set of "eyes" before I bother any real eyes.