> How useless are you for that cool startup?

> [...] after noticing how hard it is to get honest feedback when applying to a YC startup or something else entirely.

> It's a custom 5-minute challenge that roasts you after.

That's OK.

I've been working long enough in software engineering, to see the interviews turn... from collegially getting a sense of what it would be like to worth together, on the product and as part of the team... into frat hazings, negotiation negging opportunities, general corporate dysfunction, and fluffing some incumbent's ego.

So I think workers are all set, without roasting.

You know who could use a dose of humility, though?

I can't be the only one having had fun writing out some approaches and getting roasted for it?

It must be hitting a nerve if people use it and it's getting overloaded (or they're cheap w/r/t API budget).

glad you're enjoying it! we've been increasing rate limits as it gets more usage.

we definitely hit a scale larger than expected, API budget wasn't really a factor tho

OK, pretty funny.

The issue is you're given 5 minutes to write a solution, and then criticized for the lack of depth (mercilessly and hilariously, granted). This is actually great practice for interviews, but not great at determining technical depth.

At the level of "five minutes to describe the entire system and its response to the worst confounding factors imaginable" it is not fair to say "You used keywords but didn't describe in detail how ... ".

It also guessed wrong that I'd never done any of this before, probably b/c of above issues.

As a way to belittle someone who self-selected into a skills review in the middle of a workday, probably so that they'll sign up for your training curriculum, then man, I gotta say it's probably going to work well. Especially once it's not an obvious snark generator and learns to cut a little deeper.

good feedback! easy to increase the timer and support for what gets checked

initial goal was make it as easy as possible for someone to try, and in early tests, people liked the quick challenge

and hear you on misuse. initial goal was a playful way to help people looking. not designed for on the job reviews

Apparently not as useless as this website.

yeah that's my bad, my cofounder posted here I wasn't expecting it to hit the scale it did so fast

“Failed to generate challenge. Please try again.”

I'm 93% useless for having written in 5m a plan that covers all layers of failure, keeps in touch with stakeholders and would very likely lead to the resolution of the issue in <15m (nvm that I literally did this job in the past with great success).

The question was loaded as it told me that "stakeholders want to know whether it's your autoscaling script you wrote last week", it gave me the context of "alerts firing off at 2:43 am, nobody knows why" and then afterwards implied I should have replied with a very specific plan to code-review and debug my script... at 2:43 am in production with "catastrophical failures coast to coast". I have the feeling it wanted me to use all the available information to reply, rather than follow a sound plan to respond to an emergency.

Without a doubt I should have hotfixed with root cause analysis in 1m in production at 2:43 am after being thrown off the bed, and simply stared at the application recovering for the remaining 4m.

I really don't understand what's the point of this LLM-backed roaster, and if there is one, it doesn't seem to close to achieving it.

Some of these questions are odd at best. But I guess this is what I'd expect out of a recruiter who had very surface level knowledge of a subject:

Firmware Resilience for a Voice-Activated Device The home assistant prototype just hit the lab, and its latest voice command triggers a rare crash—barely reproducible, but critical. Debug traces hint at a race condition when processing real-time audio and sensor interrupts under low-power standby. Today, you examine how the interrupt service routines interact with the scheduler, ensuring audio capture stays seamless even as the device maintains privacy guarantees and maximizes battery life. Firmware must not leak sensitive audio fragments after a crash. The hardware platform is arm-cortex based and will see hundreds of millions of users relying on every subsystem working as one.

thanks for the feedback. we're trying to generate questions on the spot, and have more ways to improve here

I lead a team that is more quick reaction with same day or next day turn around times.

I often have to coordinate with multiple technical and customer facing teams.

This system does not seem to evaluate rapid situations like that very well. I was surprised by the results.

I'm wondering what's the objective?

I typed in a bit of an answer, there's not a whole lot to go on, there's no room for me to ask questions about this scenario or get data.

When I filled out an answer it just told me I was useless. I don't really get the point. What was I supposed to learn?

The objective is to get you to sign up for their LLM-driven training so that you'll be less useless. They want you to feel like they understand something and you do not.

we don't do LLM driven training, we are working on an ai-driven platform for tailored learning but I wouldn't call that training more of a replacement for traditional textbooks with a focus on DIY project based learning

Sounds predatory and deceptive, like a cult.

Which bit, LLMs?

this is our first launch like this, so our main objective is to see what people think

why 5min? in early tests with users people liked the shorter challenges than the longer ones

good feedback for more depth and more learning interaction.

this is v1 and definitely more areas to improve to make it less useless

It fails to generate a challenge for technical writing roles. Is this aimed only at engineers?

They’re probably out of API credits.

Completely broken.

Useless even.

I always get "failed to create challenge", even if I used the placeholder example

It's hitting a rate limit somewhere - lots of 429 responses.

There's something instantly and uniquely recognizable about the tone of an AI asked to be nasty, like the world's most gently complaisant performer doing their eager best to play the role of a sadist, or a Labrador retriever pressed into the title role in Cujo. At least the ad is honest. Good luck getting enough of a response out of YC's increasingly vestigial farm league here to pitch some LP on actual funding.

Understandable - the roasting is meant to be light hearted, we thought it was funny as it roasted us, and most of the people we've shared it with thought it was cute and fun as well. We are also a non-profit, we are not looking to raise VC

You did ask for an example of what you gave every impression of striving to accomplish. I'm glad to have helped you recognize the need to pivot your growth strategy.

I got 99/100

This could actually be a useful tool - I regularly do loops of "critique this design" via AI and find it immensely useful, but you're being disingenuous if you're serious that you built this to address getting "honest feedback". I guess you are trying to be edgy, but really this is just a bad attempt for some viral marketing. I'm also fully aware that developer rage baiting was probably half the goal too, and I'm falling for it.

that is our main goal. how to get this to be more honest and supportive in specific roles/projects

goal was not rage bait. actually trying to find ways to build something useful

thanks for the feedback!

Some of this has been mentioned so far but from my side I'd say the 5 minute timer yet very complex scenario is something that sets the "candidate" up to fail immediately, certainly if they're typing all this out. You're lulled into trying to cover the big picture but needing sufficient detail (it's not clear how low to go) to make sense. Having a multi-step process where it's progressively more low level as you drill in would be great. When we do interviews we tend to do very high level boxes then drill into increasing detail covering edge cases. This is hard to do in a one shot response.