💯 You MUST use AI during interviews 💯

# [ $davids.sh ] · message #284

• @ [ $davids.sh ] · # 1822

  Instead of fighting AI with the phrase: "you can't tell what a candidate knows with AI" – the question should be: "How can we ensure we understand what they know, and how they work with AI?"

  Because literally everyone on the team will use AI every day at work.

  Want to give them a problematic piece of code and ensure they know something themselves? First, ask them to make assumptions about what's wrong in different places to hear what they know or don't know, and then see how they solve the issue in cooperation with AI.

  Want them to write some algorithm? Have them write comments about what they plan to do and in what sequence, then let them use AI to quickly flesh out each point and ask it various questions like "what else would it improve," and then accept or reject the AI's suggestions.

  Want to check knowledge of infrastructure / databases / etc.? Give them a task to write the basis of an SQL query, then see how they complete it with AI, then clarify what will roughly happen at the database level and what other options there are for improving / changing the schema.

  Want to know how someone can describe architecture? See if they open Copilot and start gradually describing each module by asking it questions.

  No, AI won't provide answers to all questions; it gets confused very easily, even with a well-crafted prompt, but it will definitely be a convenient assistant line by line.

  Who do you want to hire more?

  • A dumb developer with AI
  • A smart developer without AI
  • A smart developer with AI

  Then why aren't you testing their knowledge and ability to work with AI?

• @ Daniil Tarasenko · # 1826

  Some organizations, especially state-owned ones, are now going in the opposite direction, using AI auto-checking for the work they are provided. It's good that this doesn't apply to code yet =) Especially considering there's no reliable way to be 100% sure these checking tools work properly.

• @ [ $davids.sh ] · # 1827

  As they say: "If a cow is heading in one direction, run in the opposite."

• @ unknownfix · # 1958

  Hi, do you have any slides on which AI coding use cases are good right now and which are not? People at work are going crazy, and I need to somehow convey to management that it's not a silver bullet.

• @ [ $davids.sh ] · # 1960

  Honestly, I'm having a Copilot session with Claude 3.7 / ChatGPT 4.1, and it's degrading faster than I am: it constantly loses or doesn't understand the context of the question, makes mistakes with regex or standard Go methods, gets stuck in infinite loops of writing comments. As soon as you show it a technology with not much open-source code (like Go templates), it knows absolutely nothing.

  And the ultimate piece of shit is their terminal utility, which, when asked to create a zip file ignoring one folder, launches a command to reinstall Windows... on a Mac...

  In short, I'll sit down and poke around with the cursor again, but this is really not how it should be.

  And you can visualize all of this like this: take a real problem in a real codebase and instead of writing it manually, try to have the AI from the chat solve it. Record a screen video and set a timer for how many attempts it takes.

  Most likely, it will be quite illustrative.

• @ Dima Web 🌐 · # 1964

  Exactly) The other day I tried to get this scum to generate templates - the result was very bad.