STAR is what you do when you want to hire professional talkers.

Leetcode is not the only way to hire professional builders, but is much better than STAR.

I was surprised by the kind of things people openly admit to during STAR interviews. I don't enjoy doing them but they've surfaced enough red flags from candidates to save a previous employer a few bad hires.

I'd question whether it's really that effective. A dishonest candidate could craft a boring, safe scenario for a STAR interview question that paints them as an empathetic team-player, even if that's not how they've really acted in the past.

Meanwhile, a forthright, but flawed candidate who supplies real examples from their past they regret, but would like to improve on, would look like an iffy hire.

Leetcode is the only thing that scales consistently.

At small companies, a senior engineer can ask about past projects and get a sense of whether the candidate was a meaningful contributor or just along for the ride, but that's subjective and inconsistent when you have thousands of seniors.

Leetcode is dead, and it just doesn't know it yet. High quality LLMs killed it. Either proctor, in person onsite, or your hire likely cheated.

Leetcode can't die fast enough.

? Leetcode interviews are "proctored"

Obviously not well enough to stop cheaters.

I unironically refuse to go on video for my interviews because of my "lack of webcam" (which is true for my desktop). Literally never had a single interviewer stop due to it, and I've virtual onsite interviewed at most of the FAANGs.

What I mean by proctoring is really intense, invasive shit.

Leetcode exercises are pretty effective at weeding out incompetents. If they have a false positive rate, that's fine, you've got plenty of applicants. If they weed out some of the best that's also fine, because the number one concern when hiring is to find someone competent enough that they don't become an embarrassment to anyone involved.

Real world business decisions are dramatically less optimized than people imagine they are.

Like the SAT, it probably does a reasonable job of providing a repeatable/scalable floor for skills that are a proxy for, if generally not the same as, what's needed for the job. And, like the SAT, the effort to prepare for the test is externalized.

> I've never understood why FAANGs are so fond of leetcode

Is part of it the recent frenzy to avoid bias?

If you can't take into account how smart or personable the candidate "felt", maybe leetcode is what you're left with?

OpenAI chat could probably pass leetcode.

It can pass easy leetcodes. Here's a transcript of it solving an easy leetcode problem, and I also asked it for additional test cases.

https://pastebin.com/BpdcuWEM

There was another problem I tried where it gave an almost correct answer. Fixing the answer from then was probably faster than if I had to do from scratch.

Of course, the other thing is that there's a ton of effort that goes into writing that question so well. I think if jira tickets were that well written, I'd be able to go much faster too.

Past the edit window; https://leetcode.com/problems/two-sum/ is the problem I fed it.

That is O(n^2), when I believe it could have been O(n.log(n)).