I use mostly in combination vs code with chatgpt in a browser. Chatgpt is quite good to give good startup for an idea ! But coding with it is not buttersmooth, it will sometime only give you part of the solution, even after requesting the full code, it will try to modify some stuff and just break the code. I have to balance between 4o and o1-preview to make something. But I admit that AI in coding is a tremendous help !! As you said, you have to know how to read code yourself before copy/pasting stuff that it gives ... sometimes some bad suprise can occurs. My technik is to git commit as soon as my code works well before testing new code that chatgpt gives me.... I end-up with lots of commit in my logs but it works for me ^^
Cursor is having some issues. When accepting new code it sometimes deletes existing code. But it doesn't shows in the editor in red color during preview
One trick I use is to let the AI decide on the libraries, APIs etc... and it will pick the ones it knows best. Just like a human. A few days ago I've implemented an API of which I didn't even need to visit the API page at all. Don't try to fight the LLM, but go with its flow and guard the boundaries of that flow to guide it in a proper direction.
Not very true for Cursor. ;) You can add any docs to the chat context (you can add any docs by providing a link to it) and it works really fine! I was using OpenWeatherMap API which has no SDK nor types library and Cursor wrote really nice TS interfaces for the API responses, I highly recommend trying Cursor ;)
Would thorough testing (unit, integration, functional, usability, performance ) help to mitigate some of the downfalls you mentioned, like bugs? Does your course include how to harness AI to run tests?
I agree with everything you said, I think we already see the plateau, GPT did not get significantly better in the past 2 years, the difference between GPT 2 and GPT 3 is much bigger than GPT 3 vs o1/Claude. They got faster at generating responses definitely, and the context increased dramatically from the mere 2k tokens we had to now 200k and above (while this increase comes with less "accuracy" sometimes). What i see now is that the advancements in AI are not made by improving the AI itself but by improving the packaging, this is why Cursor is killing it, they do a lot of non AI stuff to make AI look smart. Copilot was the pioneer, i would say that the improvement in the quality of life that Copilot brought with the autocompletion is greater than the improvements gained by faster/bigger/new AI models, and the improvements Cursor brought to the table with copilot++ and the composer also are much larger than what the new release of o1 did. Another thing I see is that to me it seems that also AI companies like Open AI are admitting that models will not get significantly smarter than what they are now and they are setting up engineering solution to workaround the limitations like the chain of thoughts.
@@bvedantcodes Oh how horrible of me to not give more money to a $400Million dollar company. So they can give that extra money to the owners and not the actual developers. How could the developers and upper management survive without my money that I need for groceries. I'm a terrible, terrible person
@@bvedantcodes You are a thief then too for every Spotify/Netflix/Disney and any other subscription Free trial you participated in including the first 2 weeks of cursor you got. Cursor allows you to delete your account and get another free trial, it's a great feature. I am glad this multimillion dollar company gives us the chance for unlimited free trials like Spotify used to do. If they continue to give me free trials with the same email and no credit card attached, I will continue to take it until they do not. If this is stealing, then getting free samples at Costco is also stealing.
The amount of time I've spent fixing wrong AI code would've been much better spent reading docs and / or codebases. It's much better for asking general architecture / best practices questions (even though it makes mistakes there as well) .
AI doesn't write wrong code. It's the human who's prompting incorrectly. Every thinkable combination of code can be generated by an AI. We humans, have to figure out, how to make the AI produce that exact piece of code that you are aiming for. Once you master that, through a combination of prompting and custom tools, you become superman.
@@itskittyme I agree. Think about how you google things. If you search wrong, you will end up with the wrong information. Googling is a skill. Figuring out how to interact with AI to get the results you want is also a skill.
gpt can be a pain.. I'm pulling my hair out just trying to get a flowchat built in vueJS working with gpt. Its fairly complicated.. more so than it sounds... it can be amazing at things you think are difficult and incredibly bad at doing things that are or sound easier. you definitely couldnt use it at all if you knew no code.. but it can bump you up quite a few notches. But you have to ask it a lot of questions to ensure you do understand.. it cannot just build an app for you.
Never! For two reasons: I refuse to outsource my coding aptitude to a machine, thereby making myself less viable in the work force! This is baaad advice!
I use mostly in combination vs code with chatgpt in a browser. Chatgpt is quite good to give good startup for an idea ! But coding with it is not buttersmooth, it will sometime only give you part of the solution, even after requesting the full code, it will try to modify some stuff and just break the code. I have to balance between 4o and o1-preview to make something. But I admit that AI in coding is a tremendous help !! As you said, you have to know how to read code yourself before copy/pasting stuff that it gives ... sometimes some bad suprise can occurs. My technik is to git commit as soon as my code works well before testing new code that chatgpt gives me.... I end-up with lots of commit in my logs but it works for me ^^
Cursor is having some issues. When accepting new code it sometimes deletes existing code. But it doesn't shows in the editor in red color during preview
It happened to me as well, we need to double check everything
@@o_glethorpe yes. Even when checking the code didn't show in red. But removed though
I don't doubt it. It isn't even v1.0 yet and it's as good as it is.
My job consists of using 3rd party libraries, APIs, Providers, etc... AI can not solve those. I have to read documentations
One trick I use is to let the AI decide on the libraries, APIs etc... and it will pick the ones it knows best. Just like a human.
A few days ago I've implemented an API of which I didn't even need to visit the API page at all.
Don't try to fight the LLM, but go with its flow and guard the boundaries of that flow to guide it in a proper direction.
Not very true for Cursor. ;) You can add any docs to the chat context (you can add any docs by providing a link to it) and it works really fine! I was using OpenWeatherMap API which has no SDK nor types library and Cursor wrote really nice TS interfaces for the API responses, I highly recommend trying Cursor ;)
How is cursor AI for flutter?
Thank you max for the udemy link of this course. Instant buy for me :D
I use AI like a research assistant.
Would thorough testing (unit, integration, functional, usability, performance ) help to mitigate some of the downfalls you mentioned, like bugs? Does your course include how to harness AI to run tests?
I agree with everything you said, I think we already see the plateau, GPT did not get significantly better in the past 2 years, the difference between GPT 2 and GPT 3 is much bigger than GPT 3 vs o1/Claude. They got faster at generating responses definitely, and the context increased dramatically from the mere 2k tokens we had to now 200k and above (while this increase comes with less "accuracy" sometimes).
What i see now is that the advancements in AI are not made by improving the AI itself but by improving the packaging, this is why Cursor is killing it, they do a lot of non AI stuff to make AI look smart.
Copilot was the pioneer, i would say that the improvement in the quality of life that Copilot brought with the autocompletion is greater than the improvements gained by faster/bigger/new AI models, and the improvements Cursor brought to the table with copilot++ and the composer also are much larger than what the new release of o1 did.
Another thing I see is that to me it seems that also AI companies like Open AI are admitting that models will not get significantly smarter than what they are now and they are setting up engineering solution to workaround the limitations like the chain of thoughts.
What do you think of the latest updates with VS Code copying a lot of Cursor’s features?
thanks max
Hello Max,Can you please make video on backend scalability thorugh different service like kafka, rabbit queue etc?
I will be very thankful to you
I embraced cursor AI so much that I already used up all my free AI autocompletes already ;p
1. Delete your account.
2. Re-start your Trial.
3. Thank me later.
@@ItsJustShortsBro what a great way to pay back the developers who make this tool. 🤦♂
@@bvedantcodes Oh how horrible of me to not give more money to a $400Million dollar company. So they can give that extra money to the owners and not the actual developers. How could the developers and upper management survive without my money that I need for groceries. I'm a terrible, terrible person
@@ItsJustShortsBro people have sympathy for hard-working poor people, not thieves. This is a pointless argument. ✌
@@bvedantcodes You are a thief then too for every Spotify/Netflix/Disney and any other subscription Free trial you participated in including the first 2 weeks of cursor you got. Cursor allows you to delete your account and get another free trial, it's a great feature. I am glad this multimillion dollar company gives us the chance for unlimited free trials like Spotify used to do. If they continue to give me free trials with the same email and no credit card attached, I will continue to take it until they do not. If this is stealing, then getting free samples at Costco is also stealing.
The amount of time I've spent fixing wrong AI code would've been much better spent reading docs and / or codebases. It's much better for asking general architecture / best practices questions (even though it makes mistakes there as well) .
AI doesn't write wrong code. It's the human who's prompting incorrectly.
Every thinkable combination of code can be generated by an AI.
We humans, have to figure out, how to make the AI produce that exact piece of code that you are aiming for.
Once you master that, through a combination of prompting and custom tools, you become superman.
Wrong. It does produce wrong code
@@UnVFunLimited Read again what I said.
@@itskittyme I agree.
Think about how you google things. If you search wrong, you will end up with the wrong information. Googling is a skill.
Figuring out how to interact with AI to get the results you want is also a skill.
@@itskittyme I guess you've never had AI not follow instructions or completely ignore your code before.
gpt can be a pain.. I'm pulling my hair out just trying to get a flowchat built in vueJS working with gpt. Its fairly complicated.. more so than it sounds... it can be amazing at things you think are difficult and incredibly bad at doing things that are or sound easier. you definitely couldnt use it at all if you knew no code.. but it can bump you up quite a few notches. But you have to ask it a lot of questions to ensure you do understand.. it cannot just build an app for you.
Until they stop deleting existing code, nah
Never!
For two reasons: I refuse to outsource my coding aptitude to a machine, thereby making myself less viable in the work force!
This is baaad advice!
The guy is a TH-camr , apparently he has never worked with a real codebase, other than his hello world files!
it dosent make sense being a developer anymore 😢
Yes give up, let us take those easy jobs.
If AI takes your job, you deserve it. Be a better developer
Your statement is not true AI is not a replacement is only a helping tools,just learn how to use it
Make yourself better than the AI
Ai is just a tool before they take over the world,the level they can reach is based on your level, so the key factor is you.