I am tired of seeing those posts how GPT can solve programming problems or ace leetcode style interviews. And all of those are coming from people who might have never even tried seeing whether it solves them and also have very little knowledge of ML This post is coming from ML person (have not worked specifically on GPT-related stuff) who also participates in programming competitions (not a guy who solves everything in 30 mins, but in 80% of the time can solve all 4 problems during the competition). ---------------------- No, ChatGPT can't solve leetcode problems, can't pass Google L3 interview. And you can easily see this yourself. Even leetcode easy style questions which supposed to be solved in 3-5 mins. Of course if you will ask it to solve problems which existed for ages like 2 sum or find element in sorted array, then yes. But if you will try anything new, at best it will give you a very inefficient solution and quite likely with a lot of confidence will give you something very wrong. -------------------- How to verify what I wrote? Just go to any programming competition and select the last one (no do not ask the question which already have 1mln solutions on github in almost any language, select new ones where solutions are not added to the training corpus). Here I went to Leetcode and looked at Q2562-Q2565 (last competition, I solved in ~1h15m). Here are the solutions ChatGPT generated https://justpaste.it/av6nm I selected python as most of the people can easily understand it. But feel free to generate your own, stop just copying what other people have posted. -------------------- - solution 1 will pass just because the constrains are easy. This is O(n^2) solution and you will not pass this in G/Fb/Am interviews as linear is easily achievable. - solution 2 will clearly not pass the constrains. Again this is O(n^2) solution which can be written by anyone who spent 1h studying programming. And we need n log n solution. - solution 3. Again it is clear that we need O(n^2) solution and everyone can see how we can easily achieve it. The main part here is to write a linear solution. - solution 4. Does not work at all. But the chatGPT is pretty confident that that spaghetti code will work. So what do we have? In this competition it could have solved only easy question. If it would be an interview, it would have solved nothing. I have tried it on many other competitions and the situation is similar or even worse (will not compile). ------------------- I am not telling that ChatGPT is bad. It is amazing tech which does something that a year ago would be considered impossible. But I am not sure why people here can't use their brain and continue to repost the same shit again and again even without trying (you can easily retry what I wrote on competition in top-coder or anything else). Will we reach the point where generative models will solve those problems? I very strongly believe that yes. Can it solve yet another how to merge two dataframes in pandas, use twitter api to get last tweet, do something that was done million times before? Yes already right now. But this is not a point here. #leetcode #programminginterviewprep #openai #google
Now run your wall of text trough ChatGPT and ask for a summary version 👍
easiest way to verify is to ask a LC question that came out after 2021. chatgpt wasn’t trained on any data after that
Wrong. Chat GPT is a generative Algo. It doesn’t need to know the question. It learned the concept of DP not the question itself
Can you release a bot that can solve LC then? ;) walk your talk
Agree, I think it really shines in stuff like "implement heapsort in C" or something like that, which is syntactically annoying for a human but trivial for AI. Piecing it all together requires too much insight that's difficult to draw from bodies of text alone.
blasphemy!!! ban this heretic against the chat
Agree which makes it a good search engine. Not exactly a new problem solver
Bing Chat, there is a small typo in your text. Correct is to expect it to be RIGHT sometimes.
It’s not replacing our jobs but it sure is going to cut into Google search and stack overflow’s market share. It’s pretty damn good for answering questions quickly. I now use cgpt fist before going to Google.
I see current level AI on par with our outsourced Indian teams. They can do certain tasks but need to have things combed through before use. The tasks worth delegating to AI/outsourced teams are those in which the time saved by delegating is not completely offset by the time spent reviewing.
Tech Industry
6h
1583
Are tech workers as rich as they think we are?
Health & Wellness
Yesterday
2173
I hate my f***** life
Tech Industry
4h
991
How can my idiot brother who does real estate afford this
Tech Industry
Yesterday
8515
China CYBERATTACK on UK ? WTF
Tech Industry
Yesterday
929
When would tech job market return to pre-pandemic level?
I ain’t reading all of that.
Same
I know you are not. My target audience suppose to have slightly higher intellectual capacity.