Comment by FirmwareBurner
2 years ago
>I know the shape of the answer that ai provides.
Curios how you can tell so confidently considering LLMs are constantly changing and candidates can always refactor code to make it seem more human.
2 years ago
>I know the shape of the answer that ai provides.
Curios how you can tell so confidently considering LLMs are constantly changing and candidates can always refactor code to make it seem more human.
If they refactor, I probably can't tell. Sometimes they refactor a little. I once saw someone refactor it a lot but there was still a tell-tale sign - I ended up inviting the candidate for the interview as they put in a bit more effort rather than blindly pasting in the answer ChatGPT gave them.
I can't be 100% but in any given time, ChatGPT seem to produce very similar code for the same prompt. It names the function in one particular way. It adds way too many trivial comments.
There are minor variations in the wording and sometimes variables, but in general if I see certain things it's a strong signal that they used AI.
I added some additional things (see my original comment) and it correlates perfectly with my observations.