@@Milky____ Thing is the AI is trained on human code, and the AI is not really trained to give you the most secure or reliable code but rather something that either works or something that looks like it would work. Which to me just looks like the perfect target.
Yes, their job just becomes easier. They don't even need to know programming languages anymore. If companies replace their software with AI, all hackers need to do is jailbreak the AI with prompts in natural language. Oh wow, anyone can be a hacker now with DAN prompts.
In a few years, I wouldn't be surprised to see full application being mostly ai generated with almost no real code. At this point it will be easier and more reliable to just ask the ai to imitate a real software. Bonus points if the user can customize the program, add / remove functionalities, etc on the fly. That would be a paradigm shift
i'm totally with you Yannick i think that's probably why this seems so powerful to me, i think we are headed in that direction. im not sure to what extent this is happening at the current state of the project but imagine what will happen when it can use itself 😳 thanks for the insights man!!
"ai taking my jub away" are the people who cant actually code.... otherwise they would just laugh about this panic shows that you dont actually understand anything about ai,ml .. if u think chatgpt is going to write apps for you :D 2 digit iq monkeys should stay away from code
I actually believe it could be in 4 months. 4 - 12 months. Better AI makes the prompting and plug-in easier, but I believe even with open-source GPT, one can already build a sophisticated coding AI
@krisvq , @yunwang1243 , @tylerwhatsgood That would be soo fast, but not totally unrealistic! I'm already using gpt 4 for coding, it definitely understands the requirements, but struggle a little bit when it comes to coding slightly complex code. The day it can simulate the end result might be as fast as you think!
so if you installed your favorite visualization libraries like Plotly and provided csv data you could essentially have a DataVis subagent as part of a larger chatgpt ai? can you try it with one?
The idea that I'm working on is telling GPT-4 about cataclysm, and then having GPT-4 write the scaffold in cataclysm. Then, each GPT-4 function gets resolved by catacly.
This is actually crazy. I think it might be time for me to stop putting off learning Python. Would be interesting to see how quickly I could set up a Django backend with zero knowledge of the framework.
yes i would say go for it!! it’s a useful skill if for anything just to be able to keep up w the progress in this space, it’s hard to be at bleeding edge when you can’t test out the new tools and frameworks with code while they’re still new. I would start a little more simple and learn in the context of what you’ll ultimately use it for. In terms of a framework I would go with FastAPI to start since it’s being used as a backend for a lot of the AI projects these days and it’s a bit more straightforward than Django, but follow whatever interests you there is no wrong answer! Thank you and good luck and please let me know if there’s any way I can help!!🙏🙏
hey! What is your goal? I would say learning CS definitely worth it. You can learn so much online these days though so if you’re gonna do it I would try to not spend tens of thousands of $ in the process.
Chatgpt 4 is amazing at coding just from talking with it. First brainstorm the idea and tell it to come up with in outline of the logic. Then ask it how to improve the logic and once you have that figured out you can tell him to write it in python and within minutes you have a working game or app
yeah good point, never tried to build a full app w chatgpt but saw lots of newer examples w impressive results so im sure its p straightforward at this point . have u noticed if it somehow output shows u how to scaffold the project and set it up locally? thanks Edward!
@@tylerwhatsgood I'm usually brainstorming the idea with it, then I ask it to write out the logic as an outline. I edit it to do what I want as just normal English, then I tell it to write out all the functions with no code. Then I start asking it to do each function and to code it and assembly in mvs vsc. Once it completes, I will then post is as one and say to debug it. I have coded tons of stuff and had good luck with it
haha yeah sorta! like you write the pseudo-functions and it spits out the functionality. i think chatgpt can just do this but its cool that u have in the context all the python builtins and stuff. thanks Jossef!
Err... wow! I was just about to start learning the basics of Python in a 10 - 20 hour video tutorial. Now I am going to only watch the video course so I can understand what GPT4 is actually doing and so I can instruct it more efficiently to write advanced Python code. I need to know this for a new job in September. By then GPT5 will be available. I will also be doing some consulting so my interview for that will just involve me instructing GPT4/5 and telling the interviewer I don't code because its an inefficient use of my time. My real expertise is in business data analytics, that is what they actually will be pay me for - not coding. That will mean knowing what analysis is required to solve a business problem in order to instruct an Open AI tool to do it by this time next year.
hey Brian! I think you should still follow through with getting some python basics down, there is no downside. Tools like in this video are still new and experimental and I do think some elementary programming knowledge is worth it. Even when other language models come along i think you’ll find that the python skills will still be relevant and will give you an advantage. If possible try to learn it in the same context of what you’ll ultimately be doing so that it’s not boring! If you get started soon by the time September comes around you could be quite proficient. I will try to put together a AI with Python basics walkthrough at some point, a few viewers have requested it and might be useful if I at least did a quick intro so that some of the stuff in my videos makes sense to folks who are ramping up on their technical skills. Best of luck on your journey and please let me know if there’s anything I can do to help, thank you Brian!!
@@tylerwhatsgood Hi Tyler. I agree with you and will be learning Python to a certain level but not with the aim of becoming a programmer. I am a professor in Data Analytics at a business school and have decades of business experience too (I am age 59) so will only be learning Python only as an add on to that to keep my skills current. The field of Data Analytics is moving very quickly and am sure my students will be asking me in the next year how to incorporate AI tools. I don't teach programming though, that is taught by an outside online provider. My job is really to show how the tools can be used in addressing complex business problems. My students likewise are not going to be programmers so using AI to write code is something that will lower business costs and increase efficiency. EDIT: By the way the new employer rang me last night and want me to start next week. 😁
WOOHOOO! good luck at new gig, i think if you bring this kinda stuff up to students they'll love it and it'll also give em a glimpse into what's possible and inspire them to discover new tools maybe! its all so nascent but at least to me its so fascinating, thanks Brian!
I imagine that agis will create their own language that is a state of art performance and illegible for humans. So i belive the code era in some sort of way is over too...
haha i would probably agree but can't wait to see where we are in a year or two!!! i've noticed that it kinda "tests" since it like fixes the code if something goes wrong. im gonna mess around with it a little and actually try to have it generate unit tests and see how that goes, should be fun. thanks David!!
You should change the name of your channel. Or Sue Tyler the Creator for copyright infringement. I was interested in watching this video and RU-vid closed unexpectedly. When trying to find it again, looking for Tyler what's good only yielded results for a song by Tyler the Creator. Great content. Thanks
haha wow thank you for letting me know! i think he could probably afford better lawyers 😅😅 thanks for the heads up Alex! i wonder if its a good song lol
If some primitive statistical prediction model can write a code for you based on a short prompt, it only means you're using a really crappy language, and that your short prompt contained the same amount of information as the resulting code. Which means, most of the code is not information, but just a noise. Use better languages instead, Python is noisy, verbose, and overall awful. Also, a less noisy language would have been far more efficient target for a GPT.
@@tylerwhatsgood of course, I've been using Language-Oriented Programming for many years. Constructing very small eDSLs for every problem allows to express solutions in very dense and nearly-natural language, and this approach fits perfectly into the abilities of LLMs. A nice side effect of this approach is that it allows to naturally eliminate complexity - anything that can be expressed in terms of compilation can be easily deconstructed into a trivial sequence of transforms, each transform as simple as you want. And, surprisingly, pretty much everything, besides some arcane computational problems, can be expressed as compilation, and, therefore, built as a sequence of very small eDSLs.