Ask HN: How do you detect and work around candidate using ChatGPT in interviews?
By: rajasimonIn remote interviews, it's getting harder to tell if someone is answering questions themselves or relying on AI tools like ChatGPT. What are some smart ways to verify a candidate's actual ability, without making the process feel like a trap?
By: low_tech_love
1 day agoI don’t interview but I assess students regularly, and my main approach right now is to make it very clear to them that overly generic and superficial statements, like those made by LLMs, will get their grade down. If they want to pass they’ll have to give me objective straight answers with enough depth. It’s not 100% perfect but it gives me the room to do my work a bit more comfortably when I know that the rules have been shared and accepted by both sides.
By: Proofread0592
1 day agoI don't believe there is a way besides having them come in for an in-person interview. All of the interview monitoring tools can be circumvented by having anither device nearby with a friend feeding the answers.
I think the more important question is: do you allow ChatGPT use on the job? If so why would it be disallowed during the interview?
By: sherdil2022
1 day agoOr even allowing 5-10 minutes to do whatever needs to be done to come up with a discussion about the problem - what it is: assumptions, possible solutions, coding, memory/time O(n) analysis, scope for improvements etc etc.
With the way interviews are currently conducted, many (like me who get debilitating panic attacks and anxiety during interviews) can't showcase their problem solving and coding skills - because more than their software skills, these interviews/interviewers seem to be assessing anxiety!
With a few minutes alone time to gather and organize thoughts, run through an LLM, come up with effective communication and have a back-and-forth friendly discussion - like how it is with colleagues. No one would want to work in an environment where everyday is like getting through stressful situations - needlessly I should say.