GitHub Copilot CRUSHES Leetcode Interview Questions! 😳

GitHub Copilot might not be ready to take my entire job yet, but it would certainly outperform me in a FAANG style technical interview!

I tested Copilot against a variety of Leetcode problems, and it immediately solved most of them. There were a few instances where the algorithm needed a bit of tweaking and occasionally it was way off, but overall the performance was extremely impressive!

---
Join the Community:

---

Community size at time of posting:
- Subscribers: 11699
- Channel Views: 287447

[abhishek.rathore]

Now we need an AI to perform the reverse job, take in code as context and generate comments which explain the code.

[arnobchowdhury]

Github Copilot: I am coming for your job.
Me: I don’t even have a job.

[ribeiroWilliam]

programmers on AI that will take other people's jobs: "I sleep"

programmers on AI that will take their jobs: "REAL SHIT"

[endepointe]

I have a strong suspension that copilot was trained with these questions as the models. Impressive nonetheless and eventually we will have a dictionary of problems that have been properly defined by humans (hoomans)

[kushagrabainsla]

This is trained on the shitty code I wrote in my GitHub repos, nothing to worry about !!

[MachineLearningStreetTalk]

We tried it on some non-public leetcode questions, it failed miserably

[gixxerblade]

New developer posting requirements: can write sentences that harnesses copilot

[RahulChauhan-wijv]

When he said "im not even sure whats happening here" i felt that

[rumble]

The day product managers can actually write down what they want in a structured way and understand the steps to get there is the day we should be worried, until then co-pilot and AI is just another tool

[visintel]

Wouldn't be surprised if leetcode problems were part of the dataset the AI was trained on

[kaseywahl]

In other news,
Calculator CRUSHES math equations!

[michaeldausmann]

I tend to agree with the comment at the end of the video. solving neat little functions like this is like 1% of the developers job and we would usually reach for a library method or a stack overflow answer for this type of thing anyway. You don't see bricklayers re-figuring out the recipe for cement every time they mix a new batch. The other 99% of the job is figuring our what the hell the problem is, what your organisation is like, what people are passionate about. Then figuring out what is going to work as a solution for people. Then figuring out what parts of that solution have already been built and make buy vs build decisions. Then glue all this stuff together (here is where your 1% occurs, some of the glue code might have interesting problems in it) . Then test and deploy it. then answer 1000 questions about how it works or doesn't work.

[CodyEngelCodes]

Finally, I can stop grinding Leetcode questions.

[samuelmarndi]

This will be a great helping hands for Pro programmes who know exactly what they want. After that nothing can stop them

[ONDANOTA]

Programmers:"This will take our job!"
CEO of a small company :"Computer, I would like a bubbly horror interface that smells like cinnamon"
Copilot:"Sorry, Dave, I am only allowed to output code from meaningful sentences"

[Mohit-ilgx]

When This "copilot" will be "pilot" our jobs will crash 😂

[bsal]

Man, this is so impressive 🔥 I didn't even understand that permutation one and copilot solved it.Github's team must've worked real hard on this 🔥

[thenomadcoder]

i've subscribed because our jobs are safe and because of the guitars at the end, thanks for sharing!

[WhenItsHalfPastFive]

This is so scary man. I'm already struggling to find a job.

[sagjounkani]

One thing with GPT language models like github copilot is that they have a large capacity to remember the text seen during training. So it's not surprising that it performs well on pretty common interview questions it may have seen multiple times while training on the internet data.