What’s wrong with the usage of tools? Before LLMs I was searching for “behavioral interview questions” all over the net. I was searching github for previous home take assignments, I was asking ex employees about how company is like, and what kind of questions they ask in the interview. It took time. Nowadays if I can save some time using LLMs, again, what’s wrong with that?

People don’t work in isolation. We use tools to enhance our productivity. If you ask me to write code in Notepad, I won’t be as effective as I could be if i were using Jetbrains IDEs. If I need to do some calculations, I’ll rely on a calculator rather than doing maths in my head. I’ll use grammarly (well, not anymore) to polish my non-native english writing… LLMs are just tools man.

If any, you want to hire people who are good with tools, rather than “geniuses”

1. Interview candidates with cameras on.

2. Do not ask basic software literacy questions. First of all, this was completely stupid even before LLMs. Secondly, its easy to cheat. If you absolutely have to do this then do it terms of measures. Most people in software are entirely incapable of measuring anything and LLMs cannot fix their personality deficiency.

3. Ask all questions where the expected answer is a not some factoid nonsense but a decision they must make. Evaluate their answer on the grounds of risk, coverage, delivery, and performance. For example if you are interviewing a AI/ML guy ask them about how they overcome bias in the algorithms and how they weigh the consequences of different design outcomes. If they are a QA ask them about how they will take ownership of quality analysis for work already in production or how they will coach developers when communicating steps to reproduce a defect.

4. As an interviewer you should know, by now, how to listen to people. That is so much more than just audible parsing of words. If their words say one thing, but their body language says something different then they are full of shit. Its okay that they aren't experts in everything. Their honesty and humility is far more important. They can get every question wrong, but if their honesty is on and they can make solid decisions then they are at least in the top half of consideration. 5. Finally, after evaluating their decision making ability and risk analysis then ask them for a story where they have encountered such a problem in the past and had to learn from failure.

This question comes up at least once a month so this answer is copy/paste from a prior comment.

Well; anything that’s added here is immediately training food for the next model..:..:.

Just ask them how photons lose energy due to inflation and where that energy goes.

Probably best to ask them detailed questions about their particular resume. In theory you could cram all that into an LLM but the quality will certainly deteriorate.

I haven't interviewed people at all in a year or so though.

I'll ask them to sketch out a systems architecture in real time for me. Like put the camera on the paper and tell me how the data should flow, what tools are being used etc

Not a complete answer by any means, but it may be worthwhile plugging your questions into popular LLMs. Recently I interviewed a very suspiciously behaving candidate. The algorithm they approached the problem with, which I had not seen anyone use in years of asking that question, was also the exact same that ChatGPT suggested for the problem.

One suggestion I saw recently in a thread was asking deliberately incorrect questions, such as how to implement a particular solution using an irrelevant technology. LLMs are so 'eager to please' they seem to just BS some nonsense in response. Not sure how I feel about that approach however.

Nothing changes. I ask them to describe the project that they are most proud of or they found the most challenging and ask them to describe the business and technical implementation and dig into their thinking process.

I have a standard set of behavioral questions.