How do we prove we built a project without AI when AI can even generate git histories now?
Posted by 0x03B4@reddit | learnprogramming | View on Reddit | 20 comments
With AI everywhere it feels like anybody can spin up a polished project and a fake looking repo in a day. Even Git commit histories are easy to spoof now.
So my question is how do developers stand out and show that they actually built their projects themselves? Recruiters do not check deeply and AI tools can generate huge repos that look legit at a glance.
What are the real signals that show genuine work nowadays?
Curious how others are handling this new problem, especially juniors trying to prove they actually know how to build things.
Joewoof@reddit
As a teacher, I can assure you that experienced programmers can tell immediately.
That said, it doesn’t really matter as companies will do “old faithful” during the interview: coding with pen and whiteboard.
RonaldHarding@reddit
As a senior developer I'll back this up. I've had to give a lot of code review feedback to rewrite sections of code the LLM made difficult for a human to read or understand. At the end of the day, we the human developers are going to be the ones responsible for that code, so it needs to be expressed such that we are the audience.
SaltatoryImpulse@reddit
By knowing everything about the project duh!
What issues did you encounter in the beginning? How did you resolve X issue? How does it all work? What would happen if we were to replace Y with Z in function F?
Agreeable-Leek1573@reddit
But you can use an AI to help build your project and still know everything about it.
BroaxXx@reddit
No. No you won't because you didn't build it.
Agreeable-Leek1573@reddit
Apparently I hit a pretty unpopular opinion here. But it's possible to use AI to build things and to still have the same completed understanding you would if you didn't use it...
Maybe you have to learn without it to get the full learning experience. But I've done that for 20 years. When I use AI I still have a complete understanding of what I'm doing.
Rain-And-Coffee@reddit
I feel there's a difference between:
Telling an AI "Add Auth to my App", and then having no clue what it did. This is what many non-experienced people do.
Compare to saying "Use library X to add Auth, use JWT tokens, blah blah blah". In this scenario you know all the relevant details and AI is just saving you typing.
tb5841@reddit
Only if you use AI well.
SaltatoryImpulse@reddit
No one has a problem with that. Literally. Corpos want people just like that now a days.
Intelligent_Bus_4861@reddit
Why do you even need to prove that, most people that hire don't even know how to code. Just show them that you are capable of building what your Employer wants you to. It should be their responsibility to check if person who they hired has no clue what they are doing.
HasFiveVowels@reddit
If it’s a skill AI can reliably perform, then it’s not marketable
SkynetsPussy@reddit
Before AI, how did the process differentiate between people who coded their OWN projects and people who just followed a codealong?
This is not a new problem to be solved.
ValentineBlacker@reddit
Well... you could always like, fork a big project and doctor the git history. This is not new, git history is not any kind of a real record. But I don't know if they ever did more than glance at projects in the first place.
Aggressive_Ad_5454@reddit
First of all, just say so. Claim authorship. put this line in the readme. “ I, 0x03B4, wrote this code myself without using any LLM”. If you looked at other code to learn key stuff, give credit.
Second, if it comes up in an interview be prepared to answer detail-level questions about your design and coding choices.
Interviewers (usually) aren’t trying to trip you up, just to determine whether you will make a productive colleague.
LetsHaveFunBeauty@reddit
Better documentation
ShineReaper@reddit
This whole AI thing will run itself out, because humans are humans. Enough people will try to fake it with help of the AI until they make it, but in reality they're not making it and their lies at some point come to light. More and more companies test rigidly, if an employee actually knows, what they're doing or if they're AI fakers. And once that makes the rounds, AI will be relegated to the place, where it belongs: A tool to help, but no more, people will learn that you can't fake your way through life with AI.
OP, just keep doing what you're doing, even if you fail sometimes, keep coding on your own and you will automatically stand out against fakers and will have success with that.
Kpow_636@reddit
Leave spelling mistakes and bad grammar in code comments and commit messages 🤷♂️
heisthedarchness@reddit
As is usual with text generation, literate people can just tell. AI code stinks.
Unfortunately, that answer relies on the people looking at your work being competent, which is not a given. Instead, you basically have to make the assertion, and if someone challenges you, ask them to tell you what about the code makes them think it was generated.
This is not a satisfying answer, but unfortunately there won't be any escape from the generated code swamp for a couple of years. The good news is that truly terrible interviewers probably won't care, while those who do are more likely to be able to discern whether you actually know what you are talking about.
peterlinddk@reddit
The same way you'd prove that it wasn't another team that wrote the code.
The same way you'd prove that you were part of the team that wrote it.
The same way you'd prove that it wasn't the other team members who wrote it.
There's no way to prove to anyone that you were the one writing some bit of code, unless you have video of yourself writing it - and even then you could just have been re-typing it all in.
But then again - applying for jobs isn't about producing proof that'll stand up in court, it is about showing who you are and what you are capable of. And there has always been, and will always be, plenty of room for cheating. Most companies don't care if you cheated your way in - as long as you deliver on a daily basis, you'll get to keep your job :)
Wobstep@reddit
I think most people expect programmers to use LLMs at some level when coding. Real answer is that you can't prove a negative and if someone claims that you used ai, you can ask why they think that. They either can't explain or have a reason and then you know how to address your explanation of how you coded your project on your own. There are some services where you can analyze a document to get a score of how likely LLM made the document. If you get a bad score, there are other services to "humanity" the work to make it more legit but that is using an AI to avoid making it look ai so.