Hello Codeforces community,
I hope everyone is doing well! I’m exploring new ways to accelerate my learning in competitive programming, and I’m curious about your experiences and opinions on using large language models—specifically ChatGPT—as a personal coach. Before I commit to integrating it into my routine, I’d love to hear your thoughts on:
Has anyone used ChatGPT to brainstorm problem-solving strategies, generate hints, or review their code?
Did you find its guidance helpful for understanding algorithmic concepts (e.g. DP, flows, data structures), or was it sometimes misleading?
I’m looking forward to hearing honest experiences—both positive and negative. Your feedback will help me set up a healthy, productive workflow that leverages AI assistance without undermining my own growth.
Thank you in advance for sharing your answers!
Ready for the downvotes..... but still eager to hear what the community thinks!
GPT is bad for learning algorithms.
Few months ago, I finally decided to give the "learn with GPT" thing a go. I picked flow as the topic and did the usual — asked for the theory and then asked it to ask me some questions to test my knowledge.
It provided a wrong answer to one of the questions. I could identify it was wrong because I already knew the theory. I ended up teaching GPT flow.
So yeah, the suggestion still stands — do not use GPT for tasks you are not good at.
For flows it's probably bad. But i don't think it would be that bad for begginer topics considering OP is gray
"OP is gray" — is a good reason why OP shouldn't use GPT. Because being gray implies someone is (most likely) not good at algorithms.
Also, I don't understand why you need GPT for beginner topics when there are thousands of human-made resources for beginner topics.
Okay yeah my bad, I just reread the blog and I didn't like this line
"Has anyone used ChatGPT to brainstorm problem-solving strategies, generate hints, or review their code?"
I only assumed they were using chatgpt to learn theory, from the title and immediately went to the comments. Thats my fault
While i agree there are thousands of human-made resources, it is a matter of preference since you can just ask chatgpt for free while you usually can't do that on prerecorded video lectures. For me I don't like using chatgpt to learn algorithms, I can see why someone would use it.
I avoid replying or commenting on posts with my personal account, so it might seem im bullsh*tting since i'm gray on this account.
800-1400 range mostly requires ad-hoc thinking and not much algorithms (If not any). However, I have used prefix-sums, binary search and two pointers (basics) to reach pupil on my personal account and I don't think Chat-GPT would struggle or fail at explaining these topics.
Free versions are hell, yes. But almost everytime o3 or smth like that is very good in questions/answers
i think it will be good if i use it for hints, right?
thx bro for your help!
Yeah, I've actually used ChatGPT to help me on reviewing my code, the AI gives a brief rating about your code correctness and efficiency of your code (your big O time). It also gives suggestions on how to improve your code. But sometimes it is a little dumb that the AI dosen't really understand your question or your solution is like weird but got AC in codeforces, it will like mislead you to some wrong solutions. When this happens i will just go and watch Youtube video explanation for better understanding. So overall it helps you sometimes, but mainly i still read the editorial to see where i can review
For algorithms, yeah it actually sucks at explaining them, for me i read books or maybe use websites like USACO Guide to learn a bit more, so AI for me i think is not very good at algorithmic concepts if u ask me
I’m exploring new ways to accelerate my learning in competitive programming
That's it. AI can accelerate your learning greatly, it can be a great teacher
just don't lean on ai completely
I use o3 to debug my code sometimes, especially if my approach is different from the editorial's
I don't use chatgpt to learn specific topic. But because of my language, sometimes I couldn't understand the editorial. So I ask chatgpt to simulate test cases. And it did work most of the times.
Yes, i think if i use it for hints and after thinking at problem GPT will be good
I believe LLMs are very helpful for learning CP. For example, Gemini once told me that "the essence of lazy tags in segment trees is the compression of the operation sequence" — a perspective I had never considered before.