"Very stupid." How to pass an AI interview (and not get caught)—and, most importantly, should you?

We asked both sides about AI use during job interviews.

Contents

A popular channel for tech professionals in Poland, DzikPic, collected these interesting comments.

Candidates' experiences

«Now I can spot all cheaters instantly»

— I did this back in 2023. I wrote my own software.

It was early 2023, and I decided to learn about LLMs and Python. The best way to learn is either on someone else’s dime at work or by creating a pet project.

I was doing fine in my job—I held Lead and Architect positions. But throughout my career, I’ve always gone to interviews every six months. This keeps me on the cutting edge, which is crucial now. Sometimes interesting offers come along. I don’t change jobs that often—at most once every seven years, at least every three.

I wasn’t actively looking for a job then. But let’s just say things weren’t great financially at my old workplace, and people I cared about started leaving.

So, my «stack» was: a cross-platform program written in Python. When pressing specific key combinations, a window would appear where you could input context.

For example, in one input field you’d write a short context like: «I’m interviewing for such-and-such position.» This was the system prompt. And in the second input, you’d upload your CV, job requirements, etc.

You could also configure system variables like «temperature.» I believe a local Facebook library (I don’t remember exactly—I haven’t used it in a while) converted voice to text. Under the hood was another prompt: «everything you receive from the microphone is my voice. Everything you receive from the headphones is the interviewer’s voice.»

Then this audio text went through two LLMs:

• The first LLM—ChatGPT mini at that time. «Please extract the essence from this, the answer should be concise, no more than 150 words»—and within seconds it would return text so you could respond immediately.
• The same recording, but with a different prompt, went to the larger LLM (i.e., not mini ChatGPT). A more complete answer would come back, limited to 500 words.

All of this would be sent to my Telegram—on a phone propped against my monitor.

Later I added more functionality to include screen captures. For example, if someone drew a diagram or something: hold a key, mark the top right corner with the mouse, then the bottom left corner, release the key, and send it to OpenAI.

I tested this software in real interviews—«fine-tuning» it in battle conditions. This helped me bring the program to its final stage—I’m now on version 3. But I haven’t released it publicly, it just sits with me.

It didn’t help me that much. At most, the concise answers helped organize thoughts I already had. But the project performed excellently. I learned to work with LLMs, prompting, and Python.

I plan to continue going to interviews, but probably won’t use this «software» anymore. It’s been refined to perfection.

And one more thing… as an interviewer now, I can instantly spot all the «cheaters» during interviews. I’ve even sometimes spoken prompts into the microphone that would ruin a candidate’s answer.

«It was fun to see their faces»

— Once I had a series of interviews at a company where, from the very first call, the HR person—without understanding what she was asking—quizzed me with technical questions about OOP, patterns, etc.

I realized then that these folks were unreasonable, so I proceeded to the next stage using Copilot AI, which monitored the conversation and quickly provided hints.

And I wasn’t wrong. The next stage consisted of a call with team leads from several teams, each tasked with preparing tricky questions. They were competing to see who could ask something so complex that I wouldn’t know the answer.

But I «knew» everything and explained details to them: from the reactive systems in Vue and React to the peculiarities of different index types in PostgreSQL and how they work algorithmically. It was fun to see their faces.

They invited me to the next stage, but it was a live coding session lasting an entire workday. I declined to continue.

What interviewers say

Sometimes tech professionals use AI prompts during interviews—but what do the interviewers themselves think about this? Two perspectives from our readers.

«I also ask about LLM skills»

— I conduct technical interviews. Usually in pairs, sometimes alone. I hold a team lead/tech lead position.

When I first encountered <AI assistants in interviews>, my co-interviewer and I couldn’t understand: the candidate was answering all questions, but when asked where they had used the technology, they’d reply «nowhere, I just read about it.»

Out of curiosity, we started asking questions they definitely couldn’t answer. In the end, we didn’t advance them to the next stage and told them directly during the interview that their knowledge level didn’t meet company requirements. They were offended.

The second time, I recognized almost from the start that the person was using AI. By the third and subsequent interviews, it became obvious after just 1-2 questions.

Often, the chat starts generating a response, the person begins reading it, and then the answer completely changes once they start reading the updated text. Ask them to return to what they said before—they can’t. Plus, there are questions where LLMs give false answers without clarification, I collect these.

It’s a bit harder if someone handles the interview on their own or just glances at the monitor/notes, but then tries to use LLM help during live coding. But we’ve run our assignments through all the popular models and their style is immediately recognizable.

If a candidate is copying code from another device, I’ll ask them to rename a variable. Then their project won’t compile, and they won’t understand why, and will start checking without hiding it.

What do I think about the claim that «you can’t pass interviews without AI nowadays, they should instead be asking about LLM skills»? I think if the interview is conducted by an inexperienced interviewer, you might get away with it.

But if the person has experience in development and interviewing (I’ve conducted over a hundred myself, with 13 years in IT), they’ll likely figure you out, and you’ll leave a bad impression.

If someone has some experience and knowledge, I’d recommend using LLMs for practice interviews instead—to train yourself.

Even if you manage to pass the interview, you’ll likely be asked to leave during the probation period when it becomes clear your knowledge doesn’t match the position.

I also ask about LLM skills: from basic understanding of how to write prompts, to understanding how to optimize code, what tools a person uses for this, and how they use them.

«They use AI in a really dumb way»

— People often come to interviews with AI, but they use it in a really dumb way.

Most often, it looks like this: you ask a question, the person gets confused, takes a «pause» to think, and then delivers an overly detailed answer.

It’s especially funny when the transcript processes some words incorrectly, and they see something’s wrong in the answer and start to panic. Or when, for two related questions, a person first knows absolutely nothing and then suddenly knows every detail.

I interview analysts in technical interviews, and I don’t care about theory and such—I’m interested in what the person has worked with and how they’ve solved different problems.