Can Chat GPT Replace A Programmer? | I Used Chat GPT and Here are My Thoughts

[@nismo2992]

Don't mind the people commenting hateful stuff. I don't always watch all your videos but you really have a good way of teaching and the videos I have watched have all then super helpful and broken down in a great way. I really appreciate your videos. On the topic I've tried ChatGPT here and there for certain tasks and while it is a really powerful tool it does need some knowledge how to use what it spits out just as you observed. All the best 👍

[@LearnEverythingAboutDesign]

Thanks Nismo!

[@davesvo]

Thanks for doing a video on this. I think everyone will have a slightly different experience on using chat gpt for code, and I think it is more about learning how to prompt the system to get what you want. I don't agree with needing to know how to write code though, as the gpt can fix problems and make the code work, yes it takes some messing around with but I feel for guys like me who don't know how to code it gives me the tools to get it done. I have never written code before but with chat gpt I was able to write several functional pieces of code, yes they needed work and did not function right on the first go around but chat gpt was able to fix all the problems and even though i went back and forth with it a bunch of times to get it fixed it gave me a path to success that i didn't have at all before. Keep up the content as I enjoy your tutorials and they are really helpful.

[@LearnEverythingAboutDesign]

Thanks Dave! I have been working on a step by step video going through this process and I have had some success, but kind of like you said it was bits of code. I even went so far as to copy my prompts to a word document and reuse the same exact prompts and got 3 different results coded in different ways. I think there is a bit of luck involved as well. At first I thought I was just asking the question in a slightly different way, but copying the same prompt 3x into new chats should have given me similar results I would think? I am going to continue to work on making a tutorial video for this highlighting my own experiences, but its great to hear that you have had success. It seems to be a great help file resource. Taking bits of code and asking for an explanation.

[@tulio3dprint]

I don't know why the video makes the video opaque, for that I didn't put anything, the idea is to share, just like what you did I've been doing for a month, regards

[@LearnEverythingAboutDesign]

When the text wasn't like that it was super distracting. This video was meant to just offer up some bullet points on how to best use it. Sadly I have recorded hours of footage that I just don't want to release. I will have a walk through on using it soon once I feel its worth it.

[@MakeTechPtyLtd]

You should definitely try GPT4. Totally different for complex prompts. I don't bother using GPT3.5 and unfortunately most people's experience to judge ChatGPT is with v3.5 and have a different experience and opinion than I do. I've done some complex code and engineering with GPT4. You could feed in the API text for it to reference. If you need to do it in chunks, add this at the end. . Respond: "received" . And I'll paste more text, then let you know when I've finished pasting. .

[@MakeTechPtyLtd]

I just fed in the huge PaLM2Technical Report Google PDF in chunks then got it to summarise in dot points: From GPT4: - PaLM 2 is a state-of-the-art model with improved performance and reduced inference time compared to its predecessor, PaLM. - Scaling laws from Hoffmann et al. (2022) are independently verified, emphasizing the importance of growing training tokens and model parameters at the same rate. - Improved architecture and a more varied model objective contribute to significant performance gains. - Data mixture is crucial for model success, with even minor translation pairs enabling the model to compete with production translation services. - Training a smaller model with more tokens is more efficient than modifying the model's architecture for better inference efficiency. - Inference-time control tokens effectively mitigate toxic language harms, reducing toxicity probabilities in generated continuations. - Developers should review responsible development guides and tools, and conduct application-specific analysis to mitigate potential harms and biases. - Further scaling of model parameters, dataset size and quality, and improvements in architecture and objective are expected to yield continued gains in language understanding and generation.

[@LearnEverythingAboutDesign]

Great info thanks! I will certainly give GPT4 a try. after chatGPT and Bard I wasn't hopeful, but maybe GPT4 will be the difference. I think if these software companies can get these tools built into their helpfiles it would be an amazing resource.

[@mic03311]

I just successfully converted a MATLAB script (around 1000 lines) in python (using function and class etc.) with the help of chat gpt. I agree all your points, you need to keep correcting and ask different question to achieve the expected result. Still it helps speed up the conversation and more impressive, when asked explicitly, ChatGPT was able to improve it's own implementation to make it quicker to execute! Thanks

[@LearnEverythingAboutDesign]

Thanks for the info! I was able to get a few working programs about 80% of the way there, but had to code the rest on my own.

[@Protocol-X]

Honestly... thes vide was a complete waste of my time. If i wanted to watch 8 minutes of a blurred background, and you telling people what chatgpt cant do id click a video called what chatgpt is not good for. Instead of a baiting title that makes it feel like a video where you where actually going to show the process.

[@LearnEverythingAboutDesign]

Sorry you felt that way. Not sure how you watched 8 mins of a 6.5min video :) I filmed this video probably a dozen times including walking through the process but didn't feel it was actually helpful. After talking with people who have tried to use it for this I felt the most helpful thing was to provide a few tips and asked if people want to see a tutorial on it.

[@alpsalish]

Boohoo