Started to use https://pi.ai/discover for when I’m bored or want to catch up on news/learn new things. Not specifically dev focused but a nice change if you just want to vocally brainstorm. More conversational based. I recommend to have a try!
That’s so cool! A little nerve wracking too… just weird having a computer chat with you so casually lol The only ai I’ve ever played with is chatgpt. Basic Brad over here lol
I use codeium inside VS code because it’s free and has privacy policies and telemetry settings that fit with my companies requirements. Its akin to github copilot and is really good at making suggestions and making repetitive parts of coding easier (especially unit testing).
I bounce between Claude.ai, chatgpt, and bard depending on what im working on. It’s funny but I’m way more cordial with Claude because it has a name and so I treat it like a human, but chatgpt and bard just get whatever bare minimum sentence I can come up with to get the output I need. Claude also seems to be a bit better than chatgpt and bard for my use cases at least.
Thank you so much for your helpful words, I really appreciate it. I completely agree with you, it’s an exciting time with the advancements in AI tools. Currently, I am using Poe ChatGPT https://poe.com/
Hehe, I start to feel ‘old school’ already: I have been using Github Copilot in my VS Code since a few years, first Python, and now React. I learned how to ‘ask the right questions’ when I code, and consequently to better structure my code as well.
Also I am using ‘good old’ chatGPT 3.5 for suggestions, debugging, ‘explain that code’, and other coding related questions. I learned to ask ‘structured’ questions to chatGPT, so that I get the result I wish.
So I rather use AIs for precise coding questions, instead of telling me how to develop whole applications for instance. I saw that I learn much more this way, and usually when the answer from chatGPT is not working, it’s either I used the wrong terminology, I didn’t formulate my request the right way, or I have to break down my request in smaller parts. I find this process quite useful for learning good practices in programming in general
I use tabnine in vscode (its free) and it just help auto complete most of the time and helps with some hiccups i have in syntax since i am still a beginner.
any tips on how you prompt gpt.
i have been having some problems with it lately related to a project i was working with related to using algorithims to generate a terrain map in 2d. i got stuck on a problem and me and gpt basically ended up going in circles about how to solve it (which it never did)
I have been experimenting with chatGPT for while and the results really depend (I am a plus user at openAI, so I use chatGPT 3.5 and if needed, chatGPT 4).
This is based on my experience, but I see that one has to be very specific with the instructions prompted to chatGPT. I also had the case when it starts going into circle. In general it means:
the project has to be broken into smaller parts
more detailed knowledge is required, so that it’s possible to point chatGPT towards specific libraries for instance
I noticed that you have to know the workflow for what you are trying to achieve, in a quite elaborate way. Then you sort of guide chatGPT through it. That’s my idea at least so far.
I am right now trying to create the last of the solo projects from the React class using chatGPT as an assistant. For exactly to see how that goes. I can see that I have to break that up in many pieces and give precise instructions to chatGPT on for instance how I want to build the app, and how things should work together.
So it means that I do have to know how it all works, and not expect chatGPT to spit out my app with all different components, UX, styling, data extraction from API, security, etc.
For me it’s still a positive experience, but it takes a while to find out how to interact with chatGPT for coding apps or developing a tool.
For the fun: I am a Geospatial Data Engineer, so I am curious and interested to know what solution you might come up with for your terrain map in 2D.
I am interested to develop apps which have cartographic component(s), and perhaps using some ML library to create some models, and an AI interface to prompt what you want to do.
I switched from VSCode to Cursor (it is actually a fork of VSCode) and it has helped a lot to develop faster and solve problems easily.
With Cursor you can edit and write code with AI, chat with your codebase (any file in your computer), crawl custom documentation (just pass the URL of the third-party library) and ask questions, and more!
What I like the most is that you can use your own OpenAI API key to use Cursor at-cost and have more control of your expenses. You can choose between GPT-3.5 or GPT-4.
@ianfrye0818 Yes I have been using it for a while now, some years ago it was actually free!
For me it was very useful, but it was mainly for scripting in Python (work related), and not dealing with apps that have several files (main app, ‘components’), which have to work together. I am not sure how to deal with this with Copilot.
I am experimenting now with Cursor and at this point that might become the best solution for me, as it is aware of what is in the folder you are coding from. So you can say for instance 'use the function ‘myFunction’ which is in App.jsx from a component, and it will understand.
Now I am a bit clueless about how Cursor works, subscription wise, I don’t know for sure if it is connected to my chatGPT plus account. I am not subscribed to any of the Cursor plans, but it is using chatGPT 4, which I don’t think is free? I certainly have to figure this out
Interesting. good to know. It seems you can use your own OpenAI API key with Cursor. so you won’t be charged at all using Cursor editor to make OpenAI calls from the tool?
You won’t be charged at all just for using Cursor, it is free. If you want to go beyond the free version without upgrading to “PRO” version, you can enable Cursor to use your own OpenAI API key and be charged for each API call (see the image below).
