Chapters (Powered by ChapterMe) - 00:00 - Coming Up 00:51 - What Jensen Huang said about coding 01:38 - Now that computers can code, what does this mean for CS? 03:16 - How good are AI programmers right now? 05:05 - Benchmark Datasets 08:34 - Will AI replace programmers: SWE-bench 11:44 - Good ideas come from the building process 14:50 - The evolution of programming languages 17:52 - The benefits of learning to code, even if computers can do it 18:57 - Will we see more unicorns with 10 people (or fewer)? 22:36 - Patrick Collison: Leadership Transition 23:58 - A startup should be like a sports team, not a family 25:33 - Scaling Challenges 27:23 - Applying engineering problem solving to non-engineering issues 28:55 - What will happen if AI takes on more programming roles? 33:38 - Making better for a real customer 36:58 - The verdict - learn to code! 37:51 - Crafts People 38:07 - Outro
Summary by Each Current State of AI Programming - AI tools are becoming more reliable for simple tasks. - Advances like the sBench dataset have significantly improved AI programming. - GitHub Copilot and similar tools assist developers but cannot fully replace them yet. Learning to Code - Coding enhances logical thinking and problem-solving skills. - Understanding coding is crucial, even with advanced AI capabilities. Small Teams and Billion-Dollar Companies - AI could enable small teams to build successful companies. - Experienced founders often prefer smaller teams for efficiency and focus. Founders' Evolution - Successful founders transition from engineering to managing teams and processes. - Effective leadership is critical for scaling startups. Role of AI in Enhancing Founders' Capabilities - AI lowers technical barriers, making it easier to start companies. - Founders still need a strong foundation in computer science and engineering. Conclusion AI will ease the process of starting companies, likely leading to more unicorns. Skilled programmers and effective leaders will continue to be essential.
i don't think the end consumer will really be impacted by any of this, other than perhaps having more competitors vying for their money. the people who will be most affected by this will be the employees and employers. employers will be able to make more money with less cost, and employees will have a harder time finding a job
It's very important to under what Jensen said. He is NOT saying that there will be no use for Computer Science majors or Programmers in the foreseeable future. He is saying that the end game of the industry is to eliminate "Programming" as we now know it. I don't even think Jensen would proclaim to know when this will be possible. However, that is the end goal, to enable any human the ability to program anything using natural language in the same way anyone can now generate an image on demand.
@@nonefvnfvnjnjnjevjenjvonej3384 He's not sharing advice outside of his domain, he runs the leading company for GPU's and has been doing so for decades. His projections on the direction of the computing industry, an area where he and his company will go down in history are taken seriously for a reason. This is not a crypto token project, you will be exposed quickly if you don't know what you're talking about. Regarding his predictions on the future of programming, it's all backed up by what we see today in terms of generative Ai, and consider this the beta release. Humans are not programming with levers anymore, similarly I doubt anyone will be writing CSS or any of its derivatives in 50 years for web development. The rest will follow (JS, Python, insert your fav programming lang), it won't be a dramatic revelation either.
@@nonefvnfvnjnjnjevjenjvonej3384Oh, listen to this guy! He knows what he’s talking about, with a TH-cam video instead of $100B net worth, $3at AI company, and decades of C-suite experience. Listen to him!
@@sp123 Getting rid of programming is sensationalist black and white. Realistically it's more like what today takes 10 programmers will be done by 3 with the help of AI. This has already happened with the advent of software itself. What used to take 10 whatever can now he done with 3 using Excel.
Instagram only had 13 employees when they hit a $1B valuation and that was pre crazy inflation.. you guys should know this. What will be impressive w AI is a billion dollar company w 1 or 2 founders only
Did you watch the whole podcast? They mentioned Instagram and later on they switched the perspective that there might be thousands of unicorns with 1-20 employees.
That's up to the founder and whether they created something of value! Easy to start doesn't mean anything other than that technology will permeate more parts of society that normally wouldn't get technology.
@@GarryTan But don't you think all the biggest problems have already been addressed by traditional software approaches? Sure, AI might make it possible for a small team to profitably take on a challenge that previously would have been too small, but the size/value of that market hasn't changed. So why is something now going to be a billion-dollar market with AI as a potential tool, when it wasn't previously? I think it's more likely that AI will prove to be value destroying. For example if a 10-person team somehow replicates all of Salesforce and sells it for pennies, it will crush Salesforce. Sure, maybe that small team can enrich themselves personally, but the net effect will be a decrease in the value of all software companies.
Yea in the short term this may happen, but in the long term we will come up with a lot more new problems to solve. For example consider if someone invented a car and pitched it to business men for commute. First these men would use it to visit their offices in the same town. But due to the fast long distance commute possibilities these cars unlock, they can open offices in adjacent towns and expand their business.
@@gordon2766 there is a lot more to software than just the product itself, reputation and trust are hugely important which is why salesforce is so valuable. AI making the cost to develop software cheaper does not mean that the software created is any less value, it only means that the cost overhead is reduced.
I have been a follower of Garry since his beginning on TH-cam here. It's incredible that Garry, the CEO, is still part of these videos. That speaks tons about him and YC's current culture! 🙌
Yes! This is the topic I’ve been waiting to hear. Many people are saying the software engineer career will die with AI engineers emerging soon . Others are saying “The end of software” as these software products will become quick and easy to build for everyone so prices will go to zero... We need to talk about this more as I see people discussing this everywhere
Thank you for pushing back against the terrible advice. I'm completely self-taught and published apps (and started my own company). I am a sharper thinker and a better business leader for it. The only downside is that I have found coding to be like 'play time' when I was a kid - you can REALLY get lost in it. So I've become a worse husband 🤣
Hy @Sidmohan please how do I contact you or connect with you on social media, I believe lessons from your experience and journey will be invaluable to me as I’m embarking on my own and your support will be greatly appreciated
35:15 Diana tries to push Harj to talk about equal opportunity and Harj and Garry pivot to talking about how adopting Rails made them feel personally powerful TT
The question is not how to do stuff. The question is what to do. That's where the human intuition will come. There will exist a power law in ideas, which right now isn't there because if we could speed up the building process of any idea, we can quickly see if it's worth the effort or not. AI will bridge that gap
Learning Computer Science is fundamental, like learning language and biology is fundamental for us. We learn english to be able to communicate effectively with one another and biology to understand how we are composed. To create a thought or an idea and pass that to another person, language is used. Computer science, in this particular case, coding, is the direct analog. Other areas in computer science allow us to understand how computers and machines are structured. We will need to understand machines like we understand ourselves. They are after all, mirrors of us but in a different form.
Great discussion guys, thanks! I think what Jensen meant was that a formal education in computer science or programming is not necessary, but people can always learn the basics, including computational thinking. I teach digital marketing at a college and similarly, I would say that a formal education in digital marketing is no longer necessary (which is partly why I quit teaching in January 2023 after playing with ChatGPT for two weeks).
Yeah, treating a company like a sports team only benefits the founders. They want highly motivated people with common goals and the willingness to do whatever it takes, even if it means sacrificing their mental health and family time. But that only works if everyone in the company has the same power and gets an equal share when things go well. We all know that's not the case. As a regular employee, I get paid a set amount for the work I do within a specific time frame. It's simple. I'm not going above and beyond to make the founders rich. Sure, they take on a lot of risk, so it's fair for them to enjoy the success, as long as everyone's role is clear from the beginning.
This was a really insightful brainstorming session, loved it. Yep totally agree that AI would really help in quickly pushing out more ideas and proof of concepts (from 0 to 1), but you'll still need humans and programmers to scale and sustain such startups or companies.
Bigger question is how is that $1B valuation received? Another VC and ZIRP induced bubble, it can totally happen. Can there be a $1B+ company with 2 employees which is trading on NASDAQ? Much smaller chance of that happening.
How does that square with YC insistance on having co-founders ? Chat GPT has been an excellent cofounder for a solo founder like myself.. He's always available, never tires, and the probability of a dispute is zero. The perfect assistant.
I like the camera/painter metaphor. I think the next phase of this is knowing when/where to lean into the LLM vs let the human paint. More advanced tools don't typically make the previous tools obsolete. Airplanes are more advanced than cars, but we still use cars. The microwave is more advanced than the stovetop but with still cook on stoves, etc... but what we HAVE learned is when/where to use the microwave vs the stove, how best to utilize each. In many areas the human may be preferred to the LLM and vice versa.
If people don't want to hire people, make lifestyle companies. I'm not interested in investing money in companies that take and don't give. We already know the improvement to status quo will be temporary. I'm not going for it this go round. Apparently everyone realized the only problem with current growth models is employees.
you should definitely learn to program right now. they are saying AI will reduce need for coders. But think about it like this. If AI is so good at programming then the moment you make the leap to learn to program you can get up to speed 100x faster and build with it 100x faster and only really can you do this with programming today because the AI is so good. So learn to program, you can now write programs 10x faster!
I was a programmer 25 years ago. Done other stuff like consulting, presales and sales then and "lost" my developer skills. With AI I can no build almost everything from a prototype/MVP perspective with low-code/no-code tools. Luckily my partners are full-stack developers that know how to build scalable production systems. Production systems are a different beast for sure.
The technical coding side is maybe 40% at max in the success formula but without many other factors like focus, commom sense, resourcefulness I think it could be very hard to grow a company
As an aspiring non-technical founder, Find it really difficult to come with an mvp or prototype obviously. Even with most no code tools they seem kinda limiting and some are super complex and jumbled to use. I look forward to a time when I can use real language to program efficiently and accurately or upload pictures as examples to create something similar and able to change with prompts
I completely agree with the fact that AI could eventually replace junior developers in simple programming/engineering tasks. The question then is, how does one become a skilled senior engineer without ever being a junior engineer? People are going to try their best to skip the fundamental steps given that it will become automated by AI, but without the fundamentals, you cannot become truly great at something. So, how is one to progress through this journey of becoming a great engineer (which is a difficult and lengthy process), while being able to find work and opportunities to have some level of financial stability (given that it becomes automated by AI), to one day be able to do great work and achieve things that AI cannot. How are we to ensure that there is still a place for people to go through the necessary steps of becoming an expert in the field without having to go through years and years of school as there are going to be no jobs to allow you to gain and monetize on this experience and journey?
This, exactly this is the big problem I also see coming. Not just in software development, but a lot of fields. When asking AI to do something mediocre is enough a lot of current beginner roles will fall away. So you end up having operators and experts, but becoming an expert is barely possible because nobody is paying for it. We already see the same problem with bachelor alumni not being able to get jobs, because they don't have the skills you would learn on the job. Also a lot of juniors will run into this weird place were they want to become an expert, but will end up becoming just an AI operator, because to be productive they will have to overly rely on that.
The fundamentals of engineering will always be around. However, I think in my lifetime we will see the dawn of an autonomous world. I think it is good but we must work alongside LLMs and AI and most importantly focus on how we can use such tools to improve systems within our society (i.e education, healthcare, ect...)
AI programmers is an interesting topic. If the human touch remains important for user interface design, do you see AI being used mainly for prototyping? To Jared's point about the best Python programmers also know C, do you think human programmers will still be involved when things go wrong? (debugging) Down the road, if AI replaces the junior programmer role, how do juniors ever become senior? Hope you do another discussion like this.
Chapters (Powered by ChapterMe and @danecjensen) - 00:00 - AI programmers reliability, companies workforces 00:32 - Controversy over computer science education 05:16 - Deep learning started with ImageNet benchmark 05:51 - Challenges in machine learning, AlexNets impact, and AI race 08:30 - Coding as an image recognition task 10:17 - Engineering models messy real world 15:06 - Early programming languages coding, assembly, C, Python 18:13 - Learning to code makes LLMs smarter 22:38 - Patrick Hollisons transformation from intense engineer to effective leader 23:47 - Startup founder explains rejection of familylike model 26:01 - Smart founders learn effective leadership skills 27:18 - Oracle cofounder Larry Ellisons path to success 29:20 - Reduced team sizes, efficiency, Jevons paradox 32:29 - Peoples jobs should be freed up for innovation 36:33 - Growing up in Chile, YC changed my life 37:51 - YCs craftspeople build future
Interesting comparison with why natural language to SQL hasn’t taken off… I think that’s the first problem Perplexity started working on before pivoting to their current product
6:19 - Before MNNs (massive neural nets) engineers used hand coded feature extractors, SVMs, statistical modeling, Frequency Domain represntations, Wavelets, etc.
Currently I feel it is true. I always lacked the skill to code but had the ideas and business background. Yesterday I created a discord bot with Claude and Replit. If this works already today… what will be in 10 years? OFC, kids shouldn’t learn to code.
Introduction and Discussion of Jensen Huang's Statement on AI and Coding - 00:00:00 State of AI Programmers and Their Current Capabilities - 00:01:55 History and Impact of Benchmarking in AI and Programming - 00:04:06 Comparison of Image Recognition and Programming Tasks in AI - 00:08:49 Challenges of Real-World Programming vs. Idealized AI Solutions - 00:10:18 Future of AI in Programming: From Basic Tasks to Full App Development - 00:12:25 The Importance of Learning to Code Despite Advancements in AI - 00:17:55 Impact of AI on Software Company Structure and Employment - 00:19:28 The Reality of Scaling Companies with Fewer Employees - 00:22:03 Importance of Management Skills in Company Growth - 00:23:36 Challenges of Treating a Company as a Family vs. a Team - 00:24:04 The Role of Programming and AI in Enhancing Founder Capabilities - 00:27:07 The Future of Companies: More Unicorns or Fewer, Larger Companies? - 00:29:25 Historical Predictions and the Reality of Company Sizes in Tech - 00:30:15 Impact of Technological Advances on Job Roles and Company Structures - 00:31:13 The Value of Engineering Knowledge and Taste in Founding Companies - 00:36:17 Conclusion: The Importance of Learning to Code and Embracing Technology - 00:37:16
Programming also evolving & I wouldn’t be surprised if my git repo stores only my prompts. But who will run & maintain existing code. We still have mainframes & cobol.., So Jensen’s future is really far & we have a few generations ahead of us.
And software development is more than just writing code. Disappointing to hear them use these terms interchangeably, it shows their thinking is still very fuzzy on this topic.
Agreed that CS is not just about programming, however given sufficient knowledge LLMs can create efficient softwares end to end. Atleat the Current progress in AI agents says so..
@@jonathanmelhuish4530 they did state it was more than writing code when they said writing=thinking and how you discover more while in the process of writing code. i dont what the fook you were listening to.
If we get to that point in the future when any human can create software by telling AI what to build using human language - what will become of SaaS companies? Should we start betting on physical businesses and labor/skills?
Despite AI's prowess, human ingenuity remains essential for innovation. Learning to code continues to be crucial in shaping the future of technology. 💡
[Summary was made with AI] Introduction: The text delves into the state of AI programmers, discussing their reliability, potential impact on employment, and the evolution of programming technology. The conversation also touches on the role of AI in automating jobs and the future of computer science education. Summary: -AI programmers are evolving rapidly, with advancements in developing coding assistance tools and systems. -The text mentions the launch of Devon, an AI programming tool that sparked interest among tech founders. -Benchmark data sets like sbench have played a crucial role in advancing AI programming capabilities, paralleling the impact of ImageNet in the realm of deep learning. -The discussion highlights the breakthrough moment in deep learning with AlexNet, showcasing the transformative power of neural networks. -The evolving landscape of AI programming raises questions about the future balance between human programmers and AI algorithms. Conclusion: The dialogue around AI programmers emphasizes the shift towards human-like programming abilities and the transformative impact on the tech industry. As AI continues to advance, the integration of AI algorithms with human coding expertise will shape the future of programming and technological innovation.
I am not sure if the partners are overlooking this. And I haven't fully thought through the following, but couldn't a generally intelligent enough agent be able to create an initial MVP, and then use metrics and reason to refine software? Would humans input be useful if human usage data can be collected anyways? Could the AI programmer learn '"creativity"?
This sort of means that the AI would run in a feedback loop, instead of needing to be prompted by humans each time. Kind of like how a human would do it in a build, launch, learn cycle
AI looks to be taking interesting, creative, intellectually-stimulating jobs and leaving the butter-passing to us. Where are the butter-passing robots?
If AI can ever completely replace a senior developer, the job will switch to specification work. Otherwise you will always have to accept subpar results. It's already one of the biggest challenges in development work, aligning the expectation of what exactly is being build and fixing misconceptions. If AI does the coding, a human will still have to check the code to verify, if you ever wanna get decent results. Otherwise it's just going to be bug riddled code full of bad assumptions.
there is only one thing which Ai can replace and that is a systematically designed task, lucky for us the world is asymmetrical and thus most of the use cases are mutually exclusive events which don't mean much. You can get AI to write a code for you even deploy it run it without any issues and still the product wont work well because ultimately the consumer of products are humans. Only when the consumer of product will be an AI, then we will have a real fight.If an Ai is buyer and a seller by itself it has a market, only then we can see a revolution.
It's all about communicating your idea right to the AI (LLM). Non technical people or normal English cannot be used for creating softwares. I created a full stack (MERN) e-commerce store using gpt 3.5 (In pieces not in a single go and content on site was mine obviously and the architecture) . But it took me way much time reasoning and explaining and teaching the model to relate and remember the last response and had to repeat lot of things again and again. But if you can communicate the technical requirements well to the LLM you can create a real full stack app. But you still need to know to put the pieces of the puzzle together remembering the final image of the puzzle. In the end you need to be an engineer to create the software because AI is also a software and only a engineer knows the details and aware of the errors and can actually fix the broken things where there is a need. "Technical English is the key."
🙃 No there will be a Trllionaire monopoly , because once A entity figures out to use AI in any dramatic way, say write program autonomously, then in a year, it will be the monopoly. But, that would occur, each and every time a break through is made,
Don't discuss freeing people from labor without discussing supporting them. Those people buttered because they can't build. Building a tech start up is not how everyone should aim to spend their lives. This ecosystem was built by handsomely rewarding everyone: owner, stakeholder, and EMPLOYEES!!!!!
Amen. You come to the consumer and customer. Don't make me know source code. On Drawing a cat...maybe that's why Cuban bought that Draw me a cat Co. on sharktank...
You people (meaning: navelgazing expertise biased like i am probably at my field) missing the point: Ai is not better then most programmers, it's better then most people. Most people don't code. This is the case with a lot of fields ai will replace: the top tier expert is not the benchmark relevant to the real world, the average marginal utility is. Realise that the vast, vast majority of people DONT code for multi-billion dollar startups. Realise that the vast, vast majority of peopl don;t do online platforms with 10 million users. The world exists mainly of small, rather poorly run & poor quality services. Some of them have actualy ideas that work. Soon they will be able to make them work.
I'm building my application now ; i just give gpt and Claude prompts until i get the result i want. the grind is real but i can build anything i want for the most part. it just so insane; i remember always struggling with making a login system now i can do that in 15 mins or less
Most reasoning failures modes is 70% due to positional encoding(reversal curse, context utilization, maintain count state..etc) and oblique data(crazy raw data gets us this tier of downstream performance), but once data resembles complex chains of theoretical coding and sampling. Yea most of your points become invalid. I don’t believe in an AI god, but coding is something LM will be handle with superhuman proficiency compared to humans. UI/UX creativity isn’t a limitation, image and video generation is the solution. I guess unless you’re directly working on these models, you’re obtuse to the actual reality. Google organized the world’s information…but generative AI makes it contextual and actually useful. The thing is..there are no limits.
LLM's are verry bad at generating Algorithms based on your prompt (Something Unique) for this you need to be smart and LLM's are not smart. But telling LLM to build Boilerplate code for Node.js app with 1 router and 2 controllers (this task they can do good)
Typical selfish CEO ,if the world depends on NVDIA for programming then they will be in control and there won't be any creativity and innovation so I ama against people depending on AI and this companies for coding, people should learn to code and express their views and creativity without depending on anyone,learn to code and be independent
@@sumitsp01 we shouldn't rely on big companies to give us AI we should learn how to program and be independent,create our own AI not rely on NVDIA to do everything for us
You didn't get it, a company like Nvidia provide companies with high computational GPU to run their AI models, that's one of the services they are providing.., and that's why it seems other company depends on them
@@classicemmaeasy2292 I understand everything he says and I know what NVDIA does ,my argument is that we should not depend entirely on companies like NVDIA to provide us with GPUs or any AI services, people have to learn how to program and fundamentals of Computer science,I am in South Africa and we don't have NVDIA services like how other countries have so I don't have to wait for NVDIA to bring the services in South Africa so that I can use them,I or any other south African can learn and implement then we have a different version plus completion and minimization of monopoly.
Not sure of higher and higher level of abstraction programming language analogy works. There were many attempts at 4th gen languages allowing more abstraction in last 40 years, and it did not fully succeeded. Mainstream 3 gen seems to strike this sweet spot of balance between control, flexibility and abstraction you often need. Also, natural language is very vague and unspecific (if you are not German ☺), arguably biggest value of the developers is in their ability to solve and work around some of that ambiguity into precise solutions/implementation. Think about coding as a tool in a tool belt of a craftsman, knowing how to use a hammer does not make a carpenter out of that person. AI coding assistant might be giving that craftsman a 3d printer or water jet cutter.
I think you guys need a new definition of the word 'Unicorn'. Very basic mathematics, a Billion dollar 10 years ago are not the same Billion Dollars now. As much credit as you give to tech and other enabling factors which are making more and more 'unicorns' - you have to understand that by definition of being a 'Unicorn' - just gets easier and easier as time goes by. Is new Billion the old Million? Soon enough.
Computer science is not coding! Most programmers are not computer scientists. Hacking is not proper programming. Good system architects have always been rare. Coding is like laying bricks. No AI is going to replace a good system architect for the foreseeable future. Why don't you invite a few real computer scientists or more qualified people to talk about this stuff? Please let us know if you want proper clarification.
![[LIVE] BABYMONSTER(베이비몬스터) - DRIP | 두시탈출 컬투쇼](http://i.ytimg.com/vi/UG05pxLGwzc/mqdefault.jpg)
Chapters (Powered by ChapterMe) -
00:00 - Coming Up
00:51 - What Jensen Huang said about coding
01:38 - Now that computers can code, what does this mean for CS?
03:16 - How good are AI programmers right now?
05:05 - Benchmark Datasets
08:34 - Will AI replace programmers: SWE-bench
11:44 - Good ideas come from the building process
14:50 - The evolution of programming languages
17:52 - The benefits of learning to code, even if computers can do it
18:57 - Will we see more unicorns with 10 people (or fewer)?
22:36 - Patrick Collison: Leadership Transition
23:58 - A startup should be like a sports team, not a family
25:33 - Scaling Challenges
27:23 - Applying engineering problem solving to non-engineering issues
28:55 - What will happen if AI takes on more programming roles?
33:38 - Making better for a real customer
36:58 - The verdict - learn to code!
37:51 - Crafts People
38:07 - Outro
Summary by Each
Current State of AI Programming
- AI tools are becoming more reliable for simple tasks.
- Advances like the sBench dataset have significantly improved AI programming.
- GitHub Copilot and similar tools assist developers but cannot fully replace them yet.
Learning to Code
- Coding enhances logical thinking and problem-solving skills.
- Understanding coding is crucial, even with advanced AI capabilities.
Small Teams and Billion-Dollar Companies
- AI could enable small teams to build successful companies.
- Experienced founders often prefer smaller teams for efficiency and focus.
Founders' Evolution
- Successful founders transition from engineering to managing teams and processes.
- Effective leadership is critical for scaling startups.
Role of AI in Enhancing Founders' Capabilities
- AI lowers technical barriers, making it easier to start companies.
- Founders still need a strong foundation in computer science and engineering.
Conclusion
AI will ease the process of starting companies, likely leading to more unicorns.
Skilled programmers and effective leaders will continue to be essential.
Not all heroes wear capes, thank you sir.
AI company using AI to summarize a video about AI 😂😂
As a consumer, it's hope-inducing.
As an entrepreneur, it's distracting.
A 100 percent
what do you mean by distracting?
@@johndank2209 he has no idea what he means by distracting.. just trying to sound cool
what a deep, meaningful non sequitur
i don't think the end consumer will really be impacted by any of this, other than perhaps having more competitors vying for their money. the people who will be most affected by this will be the employees and employers. employers will be able to make more money with less cost, and employees will have a harder time finding a job
It's very important to under what Jensen said. He is NOT saying that there will be no use for Computer Science majors or Programmers in the foreseeable future. He is saying that the end game of the industry is to eliminate "Programming" as we now know it. I don't even think Jensen would proclaim to know when this will be possible. However, that is the end goal, to enable any human the ability to program anything using natural language in the same way anyone can now generate an image on demand.
@@nonefvnfvnjnjnjevjenjvonej3384 He's not sharing advice outside of his domain, he runs the leading company for GPU's and has been doing so for decades. His projections on the direction of the computing industry, an area where he and his company will go down in history are taken seriously for a reason. This is not a crypto token project, you will be exposed quickly if you don't know what you're talking about.
Regarding his predictions on the future of programming, it's all backed up by what we see today in terms of generative Ai, and consider this the beta release. Humans are not programming with levers anymore, similarly I doubt anyone will be writing CSS or any of its derivatives in 50 years for web development. The rest will follow (JS, Python, insert your fav programming lang), it won't be a dramatic revelation either.
We will never get rid of programming because it is high level knowledge work. AI can't do that it can only give insight and brainstorming
@@nonefvnfvnjnjnjevjenjvonej3384Oh, listen to this guy! He knows what he’s talking about, with a TH-cam video instead of $100B net worth, $3at AI company, and decades of C-suite experience. Listen to him!
@@sp123Sounds like copium
@@sp123 Getting rid of programming is sensationalist black and white. Realistically it's more like what today takes 10 programmers will be done by 3 with the help of AI. This has already happened with the advent of software itself. What used to take 10 whatever can now he done with 3 using Excel.
Instagram only had 13 employees when they hit a $1B valuation and that was pre crazy inflation.. you guys should know this. What will be impressive w AI is a billion dollar company w 1 or 2 founders only
Totally possible.
Telegram also comes to mind.
Did you watch the whole podcast? They mentioned Instagram and later on they switched the perspective that there might be thousands of unicorns with 1-20 employees.
Why would this be impressive?
If it’s so easy to start these companies, what makes you think any lasting value is being created?
Low barrier to entry does not necessarily imply low value outcomes.
That's up to the founder and whether they created something of value! Easy to start doesn't mean anything other than that technology will permeate more parts of society that normally wouldn't get technology.
@@GarryTan But don't you think all the biggest problems have already been addressed by traditional software approaches? Sure, AI might make it possible for a small team to profitably take on a challenge that previously would have been too small, but the size/value of that market hasn't changed. So why is something now going to be a billion-dollar market with AI as a potential tool, when it wasn't previously?
I think it's more likely that AI will prove to be value destroying. For example if a 10-person team somehow replicates all of Salesforce and sells it for pennies, it will crush Salesforce. Sure, maybe that small team can enrich themselves personally, but the net effect will be a decrease in the value of all software companies.
Yea in the short term this may happen, but in the long term we will come up with a lot more new problems to solve. For example consider if someone invented a car and pitched it to business men for commute. First these men would use it to visit their offices in the same town. But due to the fast long distance commute possibilities these cars unlock, they can open offices in adjacent towns and expand their business.
@@gordon2766 there is a lot more to software than just the product itself, reputation and trust are hugely important which is why salesforce is so valuable. AI making the cost to develop software cheaper does not mean that the software created is any less value, it only means that the cost overhead is reduced.
I have been a follower of Garry since his beginning on TH-cam here. It's incredible that Garry, the CEO, is still part of these videos. That speaks tons about him and YC's current culture! 🙌
Yes! This is the topic I’ve been waiting to hear. Many people are saying the software engineer career will die with AI engineers emerging soon . Others are saying “The end of software” as these software products will become quick and easy to build for everyone so prices will go to zero... We need to talk about this more as I see people discussing this everywhere
Thank you for pushing back against the terrible advice. I'm completely self-taught and published apps (and started my own company). I am a sharper thinker and a better business leader for it. The only downside is that I have found coding to be like 'play time' when I was a kid - you can REALLY get lost in it. So I've become a worse husband 🤣
Hy
@Sidmohan please how do I contact you or connect with you on social media, I believe lessons from your experience and journey will be invaluable to me as I’m embarking on my own and your support will be greatly appreciated
35:15 Diana tries to push Harj to talk about equal opportunity and Harj and Garry pivot to talking about how adopting Rails made them feel personally powerful TT
The question is not how to do stuff. The question is what to do. That's where the human intuition will come. There will exist a power law in ideas, which right now isn't there because if we could speed up the building process of any idea, we can quickly see if it's worth the effort or not. AI will bridge that gap
Learning Computer Science is fundamental, like learning language and biology is fundamental for us. We learn english to be able to communicate effectively with one another and biology to understand how we are composed. To create a thought or an idea and pass that to another person, language is used. Computer science, in this particular case, coding, is the direct analog. Other areas in computer science allow us to understand how computers and machines are structured. We will need to understand machines like we understand ourselves. They are after all, mirrors of us but in a different form.
Great discussion guys, thanks! I think what Jensen meant was that a formal education in computer science or programming is not necessary, but people can always learn the basics, including computational thinking. I teach digital marketing at a college and similarly, I would say that a formal education in digital marketing is no longer necessary (which is partly why I quit teaching in January 2023 after playing with ChatGPT for two weeks).
Is Digital marketing a big thing over there?
What Garry says between 24:00-25:24 might be the most useful thing you will ever hear in your business life.
Yeah, treating a company like a sports team only benefits the founders. They want highly motivated people with common goals and the willingness to do whatever it takes, even if it means sacrificing their mental health and family time. But that only works if everyone in the company has the same power and gets an equal share when things go well. We all know that's not the case. As a regular employee, I get paid a set amount for the work I do within a specific time frame. It's simple. I'm not going above and beyond to make the founders rich. Sure, they take on a lot of risk, so it's fair for them to enjoy the success, as long as everyone's role is clear from the beginning.
This was a really insightful brainstorming session, loved it.
Yep totally agree that AI would really help in quickly pushing out more ideas and proof of concepts (from 0 to 1),
but you'll still need humans and programmers to scale and sustain such startups or companies.
Bigger question is how is that $1B valuation received? Another VC and ZIRP induced bubble, it can totally happen. Can there be a $1B+ company with 2 employees which is trading on NASDAQ? Much smaller chance of that happening.
End Note Touched My Soul! Thanks for inspiring Wiz...😇❤🔥
How does that square with YC insistance on having co-founders ?
Chat GPT has been an excellent cofounder for a solo founder like myself.. He's always available, never tires, and the probability of a dispute is zero. The perfect assistant.
lol
Always available? I run out of credits every time I try to use it to do any real work.
I like the camera/painter metaphor. I think the next phase of this is knowing when/where to lean into the LLM vs let the human paint. More advanced tools don't typically make the previous tools obsolete. Airplanes are more advanced than cars, but we still use cars. The microwave is more advanced than the stovetop but with still cook on stoves, etc... but what we HAVE learned is when/where to use the microwave vs the stove, how best to utilize each. In many areas the human may be preferred to the LLM and vice versa.
If people don't want to hire people, make lifestyle companies. I'm not interested in investing money in companies that take and don't give. We already know the improvement to status quo will be temporary. I'm not going for it this go round. Apparently everyone realized the only problem with current growth models is employees.
you should definitely learn to program right now. they are saying AI will reduce need for coders. But think about it like this. If AI is so good at programming then the moment you make the leap to learn to program you can get up to speed 100x faster and build with it 100x faster and only really can you do this with programming today because the AI is so good. So learn to program, you can now write programs 10x faster!
I was a programmer 25 years ago. Done other stuff like consulting, presales and sales then and "lost" my developer skills. With AI I can no build almost everything from a prototype/MVP perspective with low-code/no-code tools. Luckily my partners are full-stack developers that know how to build scalable production systems. Production systems are a different beast for sure.
Which tools would you use for a consumer facing mobile app?
Not yet
Good for a prototype that will need to be re-written.
@@genericdeveloper3966 Facebook was a prototype that was re-writen
The technical coding side is maybe 40% at max in the success formula but without many other factors like focus, commom sense, resourcefulness I think it could be very hard to grow a company
As an aspiring non-technical founder, Find it really difficult to come with an mvp or prototype obviously. Even with most no code tools they seem kinda limiting and some are super complex and jumbled to use. I look forward to a time when I can use real language to program efficiently and accurately or upload pictures as examples to create something similar and able to change with prompts
I completely agree with the fact that AI could eventually replace junior developers in simple programming/engineering tasks. The question then is, how does one become a skilled senior engineer without ever being a junior engineer? People are going to try their best to skip the fundamental steps given that it will become automated by AI, but without the fundamentals, you cannot become truly great at something. So, how is one to progress through this journey of becoming a great engineer (which is a difficult and lengthy process), while being able to find work and opportunities to have some level of financial stability (given that it becomes automated by AI), to one day be able to do great work and achieve things that AI cannot. How are we to ensure that there is still a place for people to go through the necessary steps of becoming an expert in the field without having to go through years and years of school as there are going to be no jobs to allow you to gain and monetize on this experience and journey?
This, exactly this is the big problem I also see coming. Not just in software development, but a lot of fields. When asking AI to do something mediocre is enough a lot of current beginner roles will fall away. So you end up having operators and experts, but becoming an expert is barely possible because nobody is paying for it.
We already see the same problem with bachelor alumni not being able to get jobs, because they don't have the skills you would learn on the job.
Also a lot of juniors will run into this weird place were they want to become an expert, but will end up becoming just an AI operator, because to be productive they will have to overly rely on that.
The fundamentals of engineering will always be around. However, I think in my lifetime we will see the dawn of an autonomous world. I think it is good but we must work alongside LLMs and AI and most importantly focus on how we can use such tools to improve systems within our society (i.e education, healthcare, ect...)
AI programmers is an interesting topic. If the human touch remains important for user interface design, do you see AI being used mainly for prototyping? To Jared's point about the best Python programmers also know C, do you think human programmers will still be involved when things go wrong? (debugging) Down the road, if AI replaces the junior programmer role, how do juniors ever become senior? Hope you do another discussion like this.
Chapters (Powered by ChapterMe and @danecjensen) -
00:00 - AI programmers reliability, companies workforces
00:32 - Controversy over computer science education
05:16 - Deep learning started with ImageNet benchmark
05:51 - Challenges in machine learning, AlexNets impact, and AI race
08:30 - Coding as an image recognition task
10:17 - Engineering models messy real world
15:06 - Early programming languages coding, assembly, C, Python
18:13 - Learning to code makes LLMs smarter
22:38 - Patrick Hollisons transformation from intense engineer to effective leader
23:47 - Startup founder explains rejection of familylike model
26:01 - Smart founders learn effective leadership skills
27:18 - Oracle cofounder Larry Ellisons path to success
29:20 - Reduced team sizes, efficiency, Jevons paradox
32:29 - Peoples jobs should be freed up for innovation
36:33 - Growing up in Chile, YC changed my life
37:51 - YCs craftspeople build future
Interesting comparison with why natural language to SQL hasn’t taken off… I think that’s the first problem Perplexity started working on before pivoting to their current product
So what happened
They were working on a code model that just exclusively natural language and SQL?
@@dg-ov4cf Yes. That’s the first project Perplexity started working on
@@dg-ov4cf From my understanding, yes
6:19 - Before MNNs (massive neural nets) engineers used hand coded feature extractors, SVMs, statistical modeling, Frequency Domain represntations, Wavelets, etc.
Using AI to code AI.
Hey, your last video about sales was really interesting, why did you removed it ?
Hmmm
Currently I feel it is true. I always lacked the skill to code but had the ideas and business background. Yesterday I created a discord bot with Claude and Replit. If this works already today… what will be in 10 years? OFC, kids shouldn’t learn to code.
Introduction and Discussion of Jensen Huang's Statement on AI and Coding - 00:00:00
State of AI Programmers and Their Current Capabilities - 00:01:55
History and Impact of Benchmarking in AI and Programming - 00:04:06
Comparison of Image Recognition and Programming Tasks in AI - 00:08:49
Challenges of Real-World Programming vs. Idealized AI Solutions - 00:10:18
Future of AI in Programming: From Basic Tasks to Full App Development - 00:12:25
The Importance of Learning to Code Despite Advancements in AI - 00:17:55
Impact of AI on Software Company Structure and Employment - 00:19:28
The Reality of Scaling Companies with Fewer Employees - 00:22:03
Importance of Management Skills in Company Growth - 00:23:36
Challenges of Treating a Company as a Family vs. a Team - 00:24:04
The Role of Programming and AI in Enhancing Founder Capabilities - 00:27:07
The Future of Companies: More Unicorns or Fewer, Larger Companies? - 00:29:25
Historical Predictions and the Reality of Company Sizes in Tech - 00:30:15
Impact of Technological Advances on Job Roles and Company Structures - 00:31:13
The Value of Engineering Knowledge and Taste in Founding Companies - 00:36:17
Conclusion: The Importance of Learning to Code and Embracing Technology - 00:37:16
Good discussion!
Programming also evolving & I wouldn’t be surprised if my git repo stores only my prompts. But who will run & maintain existing code. We still have mainframes & cobol.., So Jensen’s future is really far & we have a few generations ahead of us.
I only store prompts in GitHub lol
Computer Science != Programming
And software development is more than just writing code. Disappointing to hear them use these terms interchangeably, it shows their thinking is still very fuzzy on this topic.
Agreed that CS is not just about programming, however given sufficient knowledge LLMs can create efficient softwares end to end.
Atleat the Current progress in AI agents says so..
@@jonathanmelhuish4530 they did state it was more than writing code when they said writing=thinking and how you discover more while in the process of writing code. i dont what the fook you were listening to.
Capitalists do not care about quality but quantity they will fire a 10x good engineer from the US for one from India as long as he is cheaper!
So all programming is cs, but not all cs is programming
This video came at the right time in my life. TL to explain, but thank you
Could you say what IA already change d deeply in the real job world instead what it will do in the future?
If we get to that point in the future when any human can create software by telling AI what to build using human language - what will become of SaaS companies? Should we start betting on physical businesses and labor/skills?
Despite AI's prowess, human ingenuity remains essential for innovation. Learning to code continues to be crucial in shaping the future of technology. 💡
I really want to work with the guy, who feels the company as family
Hey Y combinator, How long does it take to complete the review process? will you notify if it's get rejected?
Ahh yes, youtube is the perfect place to ask them a question
Yes they do! See the deadline and they always reply you by that time.
@@2pacbigiefan 🤣😘
@@xuanan1779 Okay, Thank you.
@@2pacbigiefandoesnt have to be perfect for him to give a shot
It will get there shortly. With all eyes being on AI and intelligent AI reasoning - the future is bright imo.
what are some companies ai or otherwise which have a great employee to valuation ratio?
funnily enough Nvidia is at the top.
Telegram is one of them 30 employees company
Reading music and playing an instrument also makes you smarter, FYI
[Summary was made with AI]
Introduction: The text delves into the state of AI programmers, discussing their reliability, potential impact on employment, and the evolution of programming technology. The conversation also touches on the role of AI in automating jobs and the future of computer science education.
Summary:
-AI programmers are evolving rapidly, with advancements in developing coding assistance tools and systems.
-The text mentions the launch of Devon, an AI programming tool that sparked interest among tech founders.
-Benchmark data sets like sbench have played a crucial role in advancing AI programming capabilities, paralleling the impact of ImageNet in the realm of deep learning.
-The discussion highlights the breakthrough moment in deep learning with AlexNet, showcasing the transformative power of neural networks.
-The evolving landscape of AI programming raises questions about the future balance between human programmers and AI algorithms.
Conclusion: The dialogue around AI programmers emphasizes the shift towards human-like programming abilities and the transformative impact on the tech industry. As AI continues to advance, the integration of AI algorithms with human coding expertise will shape the future of programming and technological innovation.
Binance и его успехи вдохновляют! Жду новых видео
Guess we will find out in a couple of years😊
so much to learn
I am not sure if the partners are overlooking this. And I haven't fully thought through the following, but couldn't a generally intelligent enough agent be able to create an initial MVP, and then use metrics and reason to refine software? Would humans input be useful if human usage data can be collected anyways? Could the AI programmer learn '"creativity"?
This sort of means that the AI would run in a feedback loop, instead of needing to be prompted by humans each time. Kind of like how a human would do it in a build, launch, learn cycle
I am one of these people coding full programs with zero coding knowledge. Your dreams are possible, don’t let them tell you otherwise.
How?
Can you explain more?
How many users do you have? Are they paying for the product?
Oo that's very cool , I would like to know more about this !
Which website or resource did you use to learn to code?
AI looks to be taking interesting, creative, intellectually-stimulating jobs and leaving the butter-passing to us. Where are the butter-passing robots?
If AI can ever completely replace a senior developer, the job will switch to specification work. Otherwise you will always have to accept subpar results. It's already one of the biggest challenges in development work, aligning the expectation of what exactly is being build and fixing misconceptions. If AI does the coding, a human will still have to check the code to verify, if you ever wanna get decent results. Otherwise it's just going to be bug riddled code full of bad assumptions.
there is only one thing which Ai can replace and that is a systematically designed task, lucky for us the world is asymmetrical and thus most of the use cases are mutually exclusive events which don't mean much.
You can get AI to write a code for you even deploy it run it without any issues and still the product wont work well because ultimately the consumer of products are humans.
Only when the consumer of product will be an AI, then we will have a real fight.If an Ai is buyer and a seller by itself it has a market, only then we can see a revolution.
It has empowered generalists in my opinion
It's all about communicating your idea right to the AI (LLM). Non technical people or normal English cannot be used for creating softwares. I created a full stack (MERN) e-commerce store using gpt 3.5 (In pieces not in a single go and content on site was mine obviously and the architecture) . But it took me way much time reasoning and explaining and teaching the model to relate and remember the last response and had to repeat lot of things again and again. But if you can communicate the technical requirements well to the LLM you can create a real full stack app. But you still need to know to put the pieces of the puzzle together remembering the final image of the puzzle. In the end you need to be an engineer to create the software because AI is also a software and only a engineer knows the details and aware of the errors and can actually fix the broken things where there is a need. "Technical English is the key."
"the process of writing is thinking."
Quotable
wr,say, can say x etc any nmw s perfx, doesnt matter
English to SQL is solved for the most part, AFAIK.
What about software becomes ubiquitous and not as monetarily valuable?
Stellar!
Same pattern is observed with "BIG DATA", what happened was "big promises with no tangible results". This a flies to the light cliche.
10 People + AI = Billion Dollar Company? This is the essence of AI future
🙃 No there will be a Trllionaire monopoly , because once A entity figures out to use AI in any dramatic way, say write program autonomously, then in a year, it will be the monopoly.
But, that would occur, each and every time a break through is made,
sometimes it’s easier to train humans than to train LLMs
Great video.
Great topic!!! Keep these coming Gary
Wow, just wow,
Don't discuss freeing people from labor without discussing supporting them. Those people buttered because they can't build. Building a tech start up is not how everyone should aim to spend their lives. This ecosystem was built by handsomely rewarding everyone: owner, stakeholder, and EMPLOYEES!!!!!
1 person + AI = Trilion Zimbabwe Dollar Company
really not fair to be super smart in the modern age
Amen. You come to the consumer and customer. Don't make me know source code. On Drawing a cat...maybe that's why Cuban bought that Draw me a cat Co. on sharktank...
AI replacing programmers is as far as it can be. Any senior, experienced programmer will tell you that.
LOVE YOU ALL
You people (meaning: navelgazing expertise biased like i am probably at my field) missing the point:
Ai is not better then most programmers, it's better then most people. Most people don't code.
This is the case with a lot of fields ai will replace: the top tier expert is not the benchmark relevant to the real world, the average marginal utility is.
Realise that the vast, vast majority of people DONT code for multi-billion dollar startups. Realise that the vast, vast majority of peopl don;t do online platforms with 10 million users.
The world exists mainly of small, rather poorly run & poor quality services. Some of them have actualy ideas that work. Soon they will be able to make them work.
I'm building my application now ; i just give gpt and Claude prompts until i get the result i want. the grind is real but i can build anything i want for the most part. it just so insane; i remember always struggling with making a login system now i can do that in 15 mins or less
No craft in copy pasting (usually)
Most reasoning failures modes is 70% due to positional encoding(reversal curse, context utilization, maintain count state..etc) and oblique data(crazy raw data gets us this tier of downstream performance), but once data resembles complex chains of theoretical coding and sampling.
Yea most of your points become invalid. I don’t believe in an AI god, but coding is something LM will be handle with superhuman proficiency compared to humans. UI/UX creativity isn’t a limitation, image and video generation is the solution.
I guess unless you’re directly working on these models, you’re obtuse to the actual reality.
Google organized the world’s information…but generative AI makes it contextual and actually useful.
The thing is..there are no limits.
LLM's are verry bad at generating Algorithms based on your prompt (Something Unique) for this you need to be smart and LLM's are not smart. But telling LLM to build Boilerplate code for Node.js app with 1 router and 2 controllers (this task they can do good)
Typical selfish CEO ,if the world depends on NVDIA for programming then they will be in control and there won't be any creativity and innovation so I ama against people depending on AI and this companies for coding, people should learn to code and express their views and creativity without depending on anyone,learn to code and be independent
But not using AI will hamper productivity. If we need to keep our pace up with others, then can’t ignore AI ( LLMs)
@@sumitsp01 we shouldn't rely on big companies to give us AI we should learn how to program and be independent,create our own AI not rely on NVDIA to do everything for us
@@webstervieira3264 agreed.
That’s why I love huggingface.
You didn't get it, a company like Nvidia provide companies with high computational GPU to run their AI models, that's one of the services they are providing.., and that's why it seems other company depends on them
@@classicemmaeasy2292 I understand everything he says and I know what NVDIA does ,my argument is that we should not depend entirely on companies like NVDIA to provide us with GPUs or any AI services, people have to learn how to program and fundamentals of Computer science,I am in South Africa and we don't have NVDIA services like how other countries have so I don't have to wait for NVDIA to bring the services in South Africa so that I can use them,I or any other south African can learn and implement then we have a different version plus completion and minimization of monopoly.
There's some absolute bad ideas coming out of YC at the moment 😂
Martin Angela Hernandez Margaret Lopez Gary
Not sure of higher and higher level of abstraction programming language analogy works. There were many attempts at 4th gen languages allowing more abstraction in last 40 years, and it did not fully succeeded. Mainstream 3 gen seems to strike this sweet spot of balance between control, flexibility and abstraction you often need.
Also, natural language is very vague and unspecific (if you are not German ☺), arguably biggest value of the developers is in their ability to solve and work around some of that ambiguity into precise solutions/implementation. Think about coding as a tool in a tool belt of a craftsman, knowing how to use a hammer does not make a carpenter out of that person. AI coding assistant might be giving that craftsman a 3d printer or water jet cutter.
too many gpt wrapper companies
too many mysql wrapper companies ... said nobody ever
@@ycombinator MySQL can’t configure itself. It’s fundamentally different.
I think you guys need a new definition of the word 'Unicorn'. Very basic mathematics, a Billion dollar 10 years ago are not the same Billion Dollars now. As much credit as you give to tech and other enabling factors which are making more and more 'unicorns' - you have to understand that by definition of being a 'Unicorn' - just gets easier and easier as time goes by. Is new Billion the old Million? Soon enough.
Like when human invented calculators they stop learning maths! 🤦🏾♂️🤦🏾♂️🤦🏾♂️
Awesome 👏
"AI worker for X" is the new tar pit
Can someone help me tap in that capital, I need funding.
talk about blockchain as well
Walker Michelle Thompson Angela Jones Mark
Computer science is not coding! Most programmers are not computer scientists.
Hacking is not proper programming.
Good system architects have always been rare. Coding is like laying bricks.
No AI is going to replace a good system architect for the foreseeable future.
Why don't you invite a few real computer scientists or more qualified people to talk about this stuff?
Please let us know if you want proper clarification.
today Sam altman say , 1 person bulid 1 billion Dollar Company.
I like
👌🏿!!!!
Start a podcast
Seems more interesting 😂
Miller Jessica Williams Elizabeth Jackson Brian
smart broad
This is a Ai Generated video