Had to like after 15:30, definitely a gem of a channel in the rough of overhyped daily video channels about AI. Hoping your channels blows up at some point, your content is high quality.
Back in the 90s I read an interview with an AI researcher who believed that coding as a job would be made obsolete by AI, however he predicted that the role of a developer will transform into something closer to a gardener or a cultivator of software. As a developer and a company owner and investor, I think we will always need people that know how to drive the best possible use of technology to drive the best outcomes.
Practical. Honest. Fantastic message. Personally, I have not been this excited about development since I first started 20 years ago. I'm having a ball with these assistants!
Good video. I think the key thing to understand for people who are concerned is this: your career and job safety is dependent on how productive you are. If you produce value (engineers) or produce the potential for future value (researchers) you will do well. If AI systems evolve to become more powerful, great, use them and become good at using them to create more value. If you evolve with the tech (something engineers know well how to do) you will probably be fine. Don't sweat it, enjoy life and work hard and learn. Always learn and strive to become more productive by learning.
my biggest issue right now is you can't take a loan for a house since you can't be assured that you'll have work / revenues in 2-8 years.... and if you loose your job, then all you did was to pay interest and they'll take the house back. we're stuck until we know what happens next like UBI, same salary? i honestly can't predict anything but i sure don't believe that "we'll just have different jobs" if AI in 36 months is AGI, then most employees are replaced by AI. Why would you ask "an ai-augmented employee" when just asking an ai to do the job is much better, faster, and less headache! so the only people with jobs will be business of 1 (one entrepreneur with a bunch of ai) and some larger business or ones where humans are desirable. But even then, who will be able to afford those services? man chaos is coming and it's stressful
Yep. Totally agree. Plus it's super fun to be working with these amazing new tools. I've got the best code tutor and reviewer and it's really accelerated my learning and enjoyment of coding.
I arrived here following Karpathy's tweet. Amazing channel! I love the auto-generated video series and can't wait to see how you utilize this framework and tools like Cursor as these technologies evolve. ❤️ I also noticed the AutoDev paper release and am curious why the Copilot team isn't integrating it into their product. Additionally, I wonder how AutoGen and AutoDev, along with guidance, might collaborate, or if there will be some cannibalization, with AutoDev potentially taking precedence, even for creating agents to accomplish tasks not specifically related to coding. Cheers
If you talk to any VC the first question after the basics will be what are you doing in the Ai space? Saying software engineers have 5 years is incredibly optimistic. I live in San Francisco and there are VCs right now trying to find companies that can create software without engineers. No one is really there yet but there is too much money pouring into replacing engineers with Ai. It’s a lot of VCs golden ticket to insane profit margins. And VCs know the first company to crack that will make billions and then all the major tech companies will follow like a tsunami. So it will come and it will come quicker and harder than it seems you’re expecting. I’m only speaking from a VC perspective and where VCs are looking to put there money now.
There is definitely a race to automate software right now. This maybe the biggest gold mine of all the AI innovations that will be unlocked. The ability to build anything, faster than anyone. This is part of my 'prompt thesis' and why we focus on agents, ai copilots, and principle led engineering so much on the channel. Soon all there will be is 'X with AI' startups because anything else will be too slow.
I completely agree with your take on this. For better or worse, as programmers, the thing we know best is - programming. So the set of tasks we will automate most rapidly using AI will be programming itself. Essentially, because we can't help pursuing interesting problems, we will innovate ourselves out of a job. At some point the leap gets made where the AI dev software can implement it's own next release. At hat point, we aren't needed anymore.
been following this channel for a bit but it's crazy that karpathy tweeted about you . I don't use Twitter, so I obviously found this news from the comments.
Please focus your coding assistant vids more on prog langs and less on web and scripters like Python and others? *_I don't hire Engineers for these anymore_* 🤷♂ CAs handle web/script langs easily, due to these langs' limits. But CAs can't yet handle complex tasks needed for writing Cuda, NumPy, TensorFlow (not using them from Py, writing them in C++) that use prog langs like C++, as these do safer multi-task, compile time compute, better mem manage, strong typing etc. *_Here_* CAs fail. I'd like CA vids tasked with e.g. fold usage in variadic templates etc, these are still needed for the industry🤷♂🙏
It's something that's going to come and it's already on the horizon in some way, and that's normal. I think part of being a good programmer/software engineer/any other position in the technology area will be knowing how to turn your knowledge into something digestive for AI, or at least knowing how to integrate your workflow with these new tools. Of course, simpler jobs with low code complexity can eventually drastically lose their value and space, passing from a team to a single person who knows how to delegate this work to AI agents who will carry it out in less time. But this is normal and you just need to make your work indispensable or irreplicable so that you don't run any risk, and this has to do not only with AI but also with any competition in the job market. (English isn't my first language). Now a question that is related to the topic but not directly, have you used Sourcegraph Cody, if so what did you think, is it better than Copilot?
the bottom line here is the following... who will you hire... Someone that never coded but is great at prompting an AI and has a wicked imagination, or a software engineer with 20 years experience that is NOT as well versed in AI? Now the thing is I cheated in that question cos I chose two people on opposite sides of a spectrum... well just move along the spectrum Devs... just move along one step, maybe two... hell I done it with Fortran to Basic then to Perl, then to Php, then to python... and all the while I always had 'language'. I don't see a new tool like AI yet coming up with the overall design for my apps, putting the whole thing together without being prompted by me...? when I as a dev having worked 40 years in everything from the birth of the web to now ... well will I know WHAT I want it to build? will I be able to describe that to the LLM? .... and remember you do not pay a mechanic for spending hours beneath your car with a wrench? No, you pay them so that when they hand you back the keys you can drive the damn car! Do you care how long they were under the car? no... and you didn't pay them for simply tightening one bolt... you paid them for knowing which of the 1000 bolts needed tightening!
Here is a bigger question; if AI combined with the power of robotics can replace every human job, what happens to humans? Additionally, what activities that were also jobs paying money (which will be a relic of the past by the way) would humans like to perform anyway in a scenario where nobody has to work? Do you think people would have a sick desire to perform accountant duties or be lawyers where crime is non-existent or would they want to pursue science and arts for they are the food our brain needs? We are an unfortunate lot in this century because we can only foresee this. In the next few centuries however this could be a reality.
I feel like it's gonna become more like other engineering profession, where u need the title to practice, less people in the field and things have more norms established, which i don't mind cause i got the title and im tired of this saturated market
As an Artist I would argue that the 'evolution' model in which change is iterative and incremental may not apply to AI. Your thumbnail image for this video is the product of a radically new way to create images, using a process that is totally novel- in effect AI Art Generators introduced an entirely new paradigm to the creation of Art. Like most people I was certain that the skills of the Artist would be incredibly difficult for machines to replicate- and that is true. But that is not what happened. AI Art Generators are not an evolutionary successsor to human artists- they are something entirely different and unforseen. They did not replace Artists, they went around them via a totally unexpected and unforseen route. The same sort of paradigm shift may also occur in terms of software development- entirely new ways of interacting with AI's to arrive at the desired outcome. Artists now find themselves competing with people who have no art skills at all, yet they weild a technology that allows them to create images that can be deemed comparable to the best human Artists. I see no fundamental reason why a similar scenario will not eventually play out in software development- a scenario in which the role of the human is to define the desired output, with most of the heavy lifiting being done by the AI. And while it will still require a human to define that desired outcome, that human could be anyone who simply has an idea they wish to implement, without really needing to understand the process by which that idea is realised in code.
I think you're right but what we might disagree on is the timeline. In it's fundamental form, images are a list of many many many numbers but still, just a list of numbers. The scenario for software engineer is MUCH more complex. Code, logic, architecture, UI, databases, infrastructure, algorithms, message brokers, and the state space that it all transforms through VASTLY outnumbers the number of states a big list of numbers (images) can be in, and therefore it's complexity and ease for an AI to replicate is VASTLY larger. Again I don't think you're wrong, but the question is when will software engineer be 50-80% automated by AI? My guess, 5-10 years.
@@indydevdan The complexity thing is interesting. From my point of view creating an image is a complex blend of understanding how light reacts to physical objects and different materials, and how that light enters and leaves the scene and is or is not reflected, refracted or absorbed . Then I must consider the psychology of colours and shape language and how these resonate in the minds of those responding to the image. If there are characters interacting in my scene I need to take into account their anatomy and a crude 'theory of mind' to ensure that their postures and expressions are correctly aligned so that the relationship between them is clear. If costumes or artifacts are present I must pay attention to their design and how that design serves the narrative of the scene I am portraying and so on. So the process is complex, even if this complexity is seldom articulated, even by artists themselves. But your point is well made, because AI Art generators do indeed seem to reduce this compexity down to a 'list of numbers', which is also really my point. Just as AI does not need to understand any of the things I described in order to create effective images, it may be that AI might not need to understand many of the things you list as currently essential to the business of creating software. What AI art generators are doing is solving the 'problem' of image creation in a way that might almost be described as 'intuitive' in the sense that they are not so much creating images but growing them- watching a complex image emerge from the intial random noise is an almost organic process, entirely different to the way that I would painstakingly build an image step by step by applying the principles described above. So it may be that in the future software will not be so much built as 'grown', emerging from the latent space of the AI's in a way not that different to way images emerge. Much of the complexity of software development-it seems to my admittedly limited understanding- is bound up with the need to make the code accessible to humans, so that humans can maintain and extend it. But code created by AI's may not need to be comprehensible to humans- as long as it performs correctly. It seems inevitable that as AI's get more advanced they will in effect create their own languages in order to create software- and this language may be totally opaque to us, but if the software works the temptation to deploy it might well overcome any doubts we may have about it's opacity. Perhaps I have taken the analogy between image creation and software creation too far here- they are as you say very different- but it strikes me that an intelligence that exists as an emergent property of code+data will soon be far more adept at creating software than we are- AI's are, after all, the ultimate 'digital natives'.
Remember back in the day there are people that will connect you if you call someone from a phone. Now that job no longer exists. A lot of jobs will be about to be eliminated in the coming years.
Code is just logic, and LLMs are getting better than human in logic. The idea is to become good at translating your idea to this logic machine so it will build the right things.