As a Software Engineer with 20+ years in the industry, working in both startups and big tech, I would love to see the face of one of these non-technical AI extremists when they realize programming is just a small part of building, operating and maintaining a successful software product (and it's usually the easiest part).
What do you think is the number one job OpenAI would benefit from automating? What skill can AI acquire that gives OpenAI an advantage over its competitors? What one skill would give OpenAI’s competitors an advantage over them? What type of expertise do you think is most abundant within OpenAI as a resource to fine tune what they are creating? When a new model comes out, what is the first thing they benchmark it on? What real world thing/tool is an AI best positioned to replace?
@@Mateobjs_ Yes. You are more than a programmer. You are a problem solver. Building and maintaining software will continue to change and become higher level. We were using punch cards in the 70s, then came higher level languages, now we have AI assistants to greatly improve productivity (which I think is a good thing), and it will be different in 10 years from now.
@@Mateobjs_ Yes. You are more than a programmer. You are a problem solver. Building and maintaining software will continue changing to become higher level and you have to adapt. We were using punch cards in the 70s, then came higher level languages like C, now we have AI assistants to boost productivity (which I think is a good thing), and it will be different in 10 years from now.
Well, that's because the previous releases we're all hype that didn't meet anyone's expectations. But having an AI that can code better than 99% of humans on the planet is a different story. Will it happen next month? Nah... But if it's at this level now, where will it be in five years? It'll be at a level we can't comprehend.
@@drwhitewashcompetitive programming problems follow a pattern there are only like a dozen kind of problems like dp ,graphs ,greedy the programmers are trained to find the pattern of the problem then the problem can be done easily
@@ceciljoel9577 exactly. It is good to know graph algorithms, greedy algorithms etc, but real world problems and their solutions are much more than just these things. Although they are definitely usable in also solving real tasks, knowing these patterns is not enough.
You REALLY aren't tuned into what's going. I build AI automations for clients...real life use cases. This will translate and AMPLIFY everything. Sure if your job is eating cheeseburgers all day, you wont be affected. But for those of us gainfully employed, this will be big.
The main point of all these AI advancements is not about tools that assist humans. They are speeding up, which replaces humans. I hope you got my point.
The job "computer" was coined in the 1600's and referred to a human profession a person who made calculations. No, we don't need human computers or human calculators. We once did, today we do not. I'm sick and tired of all this cope and the absurd linguistic arguments people make to cope. Change is happening. It is significant. Define it however you need, jobs will be lost.
Until it can read my entire codebase, analyze it, make it run more efficiently, fully compile it, and, if it fails to compile, search the code for errors, fix them, and get it to run, it's still a few years away.
Thats sound like a monolithic application . Like trying to use ink pen in space Not going to happen nor anyone have time to do it. Simple use pencil in space or microservices for same service
@@christophercarson3675in a highly theoretical case where it's only trained on specific test questions it can. That doesn't change the fact it's only infrastructure most of the tech sector can't afford nor would they have the expertise to implement in house training data without hiring a dedicated team to engineer solutions for a specific use case. It isn't as simple as just turn on and go
As a developer with some experience, I work closely with a senior and near-senior developer, focusing more on planning and some coding in the private sector. Satya Nadella put it best: don’t worry about programming disappearing-worry about software as a whole becoming obsolete. At its core, software is just a UI for a database with business logic. With Moore’s Law driving hardware efficiency and AI advancing rapidly, traditional software is evolving into AI agents and LLMs, eliminating the need for apps or typing into interfaces. Businesses, however, often resist change due to inertia-a phenomenon akin to the "frog in boiling water" analogy, where gradual changes go unnoticed until it's too late. This shift will redefine many professions, including programming, but it also opens up opportunities for those who adapt to new technologies and embrace AI-driven innovation.
Factory automation moved slower than was possible because there were sunk costs in existing factories and equipment and new equipment was only added as old equipment and processes started to need replacement. I don’t see that being a factor with AI since it allows a business to reduce facility size, close offices, and decrease head count.
Let's not compare agents with algorithms, please. Agents were designed to automate natural text processing. Like you got a tweet, you first disect it into a certain category you define. Then you realize whether it's positive or negative. If negative, you try to find out what the customer was unhappy about (again a classification task). Then use some RAG to determine additional offers for the customer and propose them. See? Natural language processing/understanding. You should definitely not use agents to do any deterministic, algorithmic task. That's what the software is for, even if it's just a GUI on top of your expensive Oracle.
I doubt that it will happen before AI can fully automate programming, project managing, and well, ALL jobs, at e level of 99 percentile, or bigger (which I think is not far actually). Big part of software actually does computationally small amount of work to make majority of user operations, like sending a message, creating account, and etc, and still needs a lot of optimization for being able to process it all. For everyday app usage consistent enough AI agents costs many orders of magnitude more than it needs to for this scenario. AI will definitely appear in any big app, for helping, or maybe performing difficult and complex operation, but it won't make software obsolete, and it is actually a very small problem, compared to AI just replacing programmers, which is also a very small problem, compared to someone creating an unaligned self sustainable, exponentially growing AI agent swarm, or super intelligent AI, using our social fragmentation, persuasion, and humanity flaws for escaping gatekeepers.
Exactly. These people would believe every word a used car dealer tells them. Sometimes I believe the man in the street has more common sense than these people.
The craziest part is programmers who said this, like they have something to gain about all this. Presumably smart people who work in IT riding the hype train is insane.
Probably a better analogy of what is coming is what happened in manufacturing. I was working in industry in the 80’s when computers and automation first started entering the factory. It wasn’t very good, expensive, and required a lot of human intervention. 30 years later I do the work of 5 machinists better and faster. But the problem AI poses for human tech workers is it’s happening basically overnight. The 4 other machinists had 30 years to retire, retrain, or just grow with the technology. There is no way AI will create more jobs than it destroys. We all can’t just sell coffee to each other.
It already happened with older AI models. The new models will soon get to the point that using a human programmer will be unsafe and undesirable not just expensive.
“All truth passes through three stages. First, it is ridiculed. Second, it is violently opposed. Third, it is accepted as being self-evident.” - Arthur Schopenhauer
OpenAI is one of the worst in terms of their sh`tty marketing. They've been overhyping releases for years. As a software engineer, I do use AI every day, in fact a lot. It does speed up the development process if you give it simple tasks to solve. It's currently on the level of a junior developer, more or less. But there's a whole world of tasks where it starts hallucinating and writing BS - mostly architectural decisions, refactoring stuff, various integrations, etc. I don't think O3 will be such a huge game changer. OpenAI can keep saying big words but synthetic AI tests are NOT the real-world tasks. And the original SWE-benchmark, where it kinda sucked, just proves this fact. I'm not in denial, but I think the truth is just between polar opinions. Is AI totally useless? ofc not. Will it completely replace all programmers? No, not in the next 5-10 years for sure. The correct answer "yes, it might take some tasks and may or may not shrink the market and cause some decrease in IT salaries, I'd say anywhere from 10% to 50%"
Don't get me wrong, I'm all for new tech and A.I. and a huge tech nerd myself, But let's face it for normal working class blue collar plebs like me which make up majority of workforce, Most of this A.I. is completely useless in real world still! They are just doing the typical sales pitch to pump up hype to get more investment money, Wake me when it's merged with a humanoid robot that can take the trash out, Mow the lawn and wash the dishes all when I ask it to in a just a few words.
I use AI for most of my issues and questions in life becuase AI is smarter and more reasonable than 99% of the cooked society we have to deal with. Like once you get used to AI dealing with people is fucking absolutely useless.
the thing is... when errors arise, who will be to blame? i think companies need someone to blame for errors, first and foremost, and it's not possible to blame a machine that doesn't care. that is, if users even care about the quality of the software they use. the direction seems to be software becomes increasingly buggy and bloated, no one to blame because it passes test, user doesn't care and will just refresh the page. be the change you wanna see, don't use bugged paid software, and go the extra mile to ensure quality in the software you produce
It's because once they say they have achieved AGI they get a huge chunk of money as part of a milestone payment that is defined in one of their contracts with Microsoft
OpenAI has specifically said they are on the path to AGI, which is not controversial. All frontier labs are on this path… whether it’s a 2 year path or a 20 year path. But OpenAI did not call o3 AGI. Many OpenAI fans and sensationalist content creators commenting on o3 are calling it that, but OpenAI is not.
openai has absolutely sh*tty marketing strategy, I simply stopped believing anything they been saying in the last years. The company itself is rotten to the core. And we all know the reason behind it
6 15 Hold on. Look at the second highest score. o3 got 76% and it looks like it cost, what, $20 per task? A human programmer (in the US) probably gets paid between $50 and $75 per hour. And the human also requires the company to pay matching contributions to 401K, some towards health insurance, holiday pay, vacation pay, etc. The AI seems pretty competitive, now.
When Orion releases next year by December 2025 and the cost to operate O3 falls below $1,000, things are going to get weird. The $1,000 price tag won't last for a year. OpenAI or some other company is going to bring the cost of an O3 model down to O1 prices.
@drwhitewash I am saying Orion combined with systems like O3 is going to cause unprecedented change to the AI industry. I also believe that Open AI and its competitors are going to push the cost of O3 systems to under $1000. Technology always gets cheaper just like computers laptops, and cell phones have done over the last 50 years AI is no different.
There will be no fixed $1000 USD price tag per task. Re the highest scoring ARC benchmarks: it was more than $1000 per task. However, the point was here to demo the power of test time compute. The pricing model will likely be based on the amount of compute you use. So you’ll be able to make it “think” shorter or longer, longer being more expensive. Regarding common programming tasks: the price tag will be a small fraction of the mentioned $1000 per task, as there’s so much more code on the internet. And huge efforts go into optimising models for coding. The model o3 wasn’t wasn’t optimised for the ARC task - somehow these LLMs and systems that utilise LLMs (such as o3) don’t natively understand 2D grids well. Basically location or movement or symmetry. Presumably something to do with lack of training data and how the tokeniser operates.
@@micahwilliams1826 no , they're tests that humans can do very easily , but that AI cannot . it's also the reason why o3 is not AGI, there are tests you can easily understand, yet o3 struggles. You can submit your own tests to ARC , I think they have a github for this
People that made the A.I., Much more complex program than all the fools pretending their job is safe at the office. Even the A.I. coders will be out of a job in coming years. lol
@@ShaneMcGrath. There is not much coding in AI at all. Many training scripts are only a couple hundred lines long. The research is the much more challenging part.
Definitely not the end of programming, Some things can only be described through programming jargon that can only be understood if you know how to program. How are you supposed to give specific instructions without knowing how to code? If you want to do anything original that is finely tuned to your requests, a hands on approach will likely be required unless the AI can read your mind.
It's heartwarming to see people who have for over a year declared that AI will never be competent at programming, now similarly declare that AI will never accomplish coding related tasks. I wonder what field they'll next declare AI will never accomplish while they're permanently unemployed with no hope of any income.
@@Justin-wj4yc They have significant limitations and make mistakes. I don't know if you're just trying to make conversation or if you honestly think this observation somehow invalidates my point.
@@kyneticist What was your point in your comment? It's like you're salty developers make more than you. AI hasn't improved much at coding in real life since ChatGPT launched. When AI is good at coding in real life not just benchmarks it will be good at everything
Why are you so salty... The point when AI actually becomes better than humans in coding is where the next big modell will be made by AI. And here is also the point where it can imporve itself over and over until all jobs that are "mind-based" are going to get replaced, exept for psychologists. (Most of "them" are employed making a lot of money btw, not sure why they would be unemployed)
AI will remove the boring routine part from coding like coding itself. Humans will be left with more challenging tasks like making coffee. These bloody Stochaistic Parrots keep getting better and better. Really annoying.
For the unemployed with ideas of creating companies this is the greatest era ever. I no longer need tutorials on UA-cam but still helpful to know the basics and lot more otherwise you won't know what to ask the ai to make you. Example of someone who knows vs someone learning and how can they can use it.... I can then ask the ai to write me a basic react components from navbars to help setting up the whole folder structure and then also and such then also add css . From there I customize based on the design I make. Today I can build full stack websites with the help of ai. Before I had to hire a programmer.... I guess you are right their days are numbered but for saas the jobs will always be there for engineers who have positions like system administrator and such positions that can't be done by ai only their agents which would need to physically be there or windows 11 will have to find a way for an ai agent to do all the administrative work with the help of 1 person
AI could never *design* systems, it can only code. Learning just coding was not enough either way. That’s why bootcamps and self-learning have low success ratio. What will change in the near future? Monkey programmers will become obsolete. Developers and software engineers that participate actively in designing the software architecture have no fear. Instead of whining about AI taking your job, upgrade yourselves from a programming monkey to a designer.
@@mo_mo_meaw I don't believe it's such a low percentage. But I don't necessarily believe it's a bad thing. As Excel rose the number of accountants, AI could rose the number of software engineers. The job market, yes, it's changing and we must conform to it, but skillfulness is NEVER wasted.
Older chatgpt models were very costly to run as well before being at commercial stage, optimisation will happen to bring prices down 10x-100x soon. They will realise eventually the agents vision, with multiple specialistic agents trained to become architects, testers, managers and collaborate all day long into software solutions. It's not here yet, but to me it's 2-3 years away to be truly job nightmare.
Pretty much, thats the real change but there's a lot of people in those places that can choose not to use it; unlike a developer who was already seen as a sunk cost.
@@hendrx well, another ignorant one that will cry in a year or two... edit: Also, o1 was revealed a couple of months ago, and they skipped o2 entirely. We will have o6 or o7 before the end of 2025 at this pace, and it will definitely be better than every human developer.
the test isn't whether it can beat programmers ; the test is can a none programmer use it to create an app to client specifications. The day a client can say to a none technical person "use your AI to create me an app to do these precise tasks and integrate it into my website" ... thats the day.. thats end of skilled dev.. but , as always I just predcit the skills will change but remain technical and difficult. Do we program in binary anymore? Majority of us not ; we use words, numbers and symbols. I think what we're seeing is programming language becoming the spoken word instead.. but that wont make it easy enough to de-skill it.
Occupations change, come and go. What the work consists of changes. Before refrigerators there were iceboxes, cabinets cooled by ice. At that time there was a market for ice delivery. That market is gone now. What developers do has changed some during the 40 years I have been working with computers. The web didn't exist in 1980. Programming languages that were common back then are not so common today; instead new languages have come. Developer environments have improved a lot. The developer job will continue to change. AI will help developers even more than today. The role of the developers might change to be more about prompt engineering. But for quite a while, I think, classical developers will still be needed, especially for those parts that AI's don't know very well.
When things break, it will be bad. Most of the time, developers spend on design, architect, integrating, testing, sustaining, communicating, and troubleshooting. Programming is only good as who cross check if it works and why it failed. If a person comes in with little programming background and develops the product. Within a few years, the complexity of maintaining that software will be tremendous. And remember, the software or AI companies will try their best to sell their product to lock in customers. But when applying to actual business cases, a lot of time, they might not meet customers' desires. Think of any software companies and their products, we want to standardize but always end up customizing the product to meet business needs. Just my 2 cents, I am also eager to see how it goes from here.
These are no longer "models". The word model has been co-opted. Instead of the model directly (in a single pass) producing an output, they are running their own agentic workflows in the backend to correct and compensate for the actual models poor performance. Hence the rising cost and long wait times. It's basically a scam. They are creating the illusion of progress by hard coding guardrails and validation environments to assess outputs before feeding them back to re-prompt the underlying model to hopefully get the right answer. These are effectively error grinding while loops. Any developer whose job was taken by AI had no business being in tech in the first place. Anyone who believes AI is driving the current job market is a mouth breather and developers who use ai code editors to generate code they don't understand have no inner monologue and don't know how they would feel if they didn't eat breakfast yesterday.
Dude you're not thinking! You're actually questioning programmer jobs and costs??? This just started! And u can already begin to see the exponentials starting everywhere. Those costs are gonna get cheaper exponentially as has been happening and YES PROGRAMNERS ARE DONE! Pretty much everybody 90% anyway is done 10 years out. In 3 years we've garnered probably 100+ years of knowledge already. We are 10xing every 3-5 months now n very very beginning n tip of the iceberg smh. People gotta wake up
@@wolfgangpreier9160 you think a programmer who builds a software couldn't use it more effective than the guys who do it now? programmers will just take their jobs because they are higher qualified.
@@wolfgangpreier9160 coincidently I am a developer working with that ai stuff. I could show you how a script learns how to use a software just by itself
@@jojojojojojojo2 Why should a programmer build a Software and then use it himself? I have done it to calculate my optimal PV and Battery size but that was for my house. When i have programmed in the past - i have stopped doing that about 5 years ago because not enough profit, it was always for a customer.
Just writing code is one thing but when it can talk to ones who ordered the code to be fixed/created, get right all the requirements and then realize some of them do not make sense, go back and talk to ordered, realize they cannot agree to anything and force a meeting with all involved parties, then make them agree on new requirements, make sure they are fine, all works and is tested and built...then a ton of jobs in AI will disappear. Coding is...coding. It is not all what programmers do. Sure many times part of all that I described is done by business analysts, testers, product owners, scrum masters etc. but still a programmer is very much involved in all of that. And that's for existing piece of software. Now do it all plus you have to do it completely from scratch and choose all the technologies and how you go about say creating a new ledger system for a bank.
As a fellow programmer myself, I've come to the conclusion that pure SWE has no future. I think it's best to veer to robotics and automation at this point. Anything that has to do with processes in the physical world cannot be automated nearly as quickly as progromming due to a lack of training data.
don't get yourself worried. the current performance of AI in coding only looks good from the perspective of Dunning-Krugers "valley of ignorance". programming consists of so much more than typing out the code. it's about understanding the problem you are trying to solve, the constraints and options you have and then tying everything together in a way that "works".
@chillgamer, Yeah, real programmers don’t do any coding. That’s just for juniors. Real programmers do grand architecture and difficult business decisions.
no, economy fucked the career in the first place, nobody that actually knows shit about llm says they are replacing softwares engineers, it's probably the end of frontends taking one week to turn a button from red to blue tho
Not yet atleast, we dont know how good it can code inside large project, even the SWE-bench test is not really clear on how much of its result are inside the training data. And these projects are all Python projects. How would it perform on c++ or rust projects where there is less training data(rust) and manuall memory managment (C++). And one pull request costing ~3k is not viable, but this will clearly decrease in the future. You should wait until o3 comes out and real testing is done
O3 is good but takes 1000$ per prompt and is its context window (how much text it can handle) is not big enough to think over an entire programming project. As programming is very intensive and demands creativity, i dont think prompting it over and over and over again is a viable option, so no nothing changes for now atleast for most companies, can't say the same about giants like google who could use it from time to time without feeling it
I believe that at least for the next years, we will still need Developers reviewing and supervising the work done by AI to make sure it is going in the right direction. But this will mean, we don't need a lot of junior devs, but instead of a team of 10 devs, we will need only one senior or software architect using AI to do the work.
May I inform you that there will be highly skill AI agents and computer use agents to collaborate with the o series models that code. This means an orchestrated AI swarm will be able to create the idea, code it, test it, send feedback, learn and revise all on its own in "2025".
I agree with you. The current systems are significantly lagging humans with general vision abilities, but this gap will also be closed soon. Massive context windows with nearly perfect recall are also just around the corner. This thing is moving at lightning speed. Altman had indicated that the o-series will be updates every few months or so. He’s said “we’ve got the next 2 years in the bag” or something like this. That takes us to o-10…
@@Justashortcomment Programmers need to look into being "Manager Engineers".... meaning how to manage these AI agents, requesting prompts, review board and how to comment on AI agents feedback of their programing skills, etc. If it is not the desired outcome of your project, do you blame the AI agent or your own prompting skills? Managing AI agents for coding will be the new skillset I feel.
@@aziz9488Construction work? When just thirty to forty percent of cognitive workers are unemployed, how much construction work do you think there will be? The economy won’t be booming, that’s for certain. When that many people lose their jobs, we’re talking millions of people globally defaulting on mortgages, credit cards, car loans, etc. leading to a banking crisis, massive drops in tax revenue , a consumption collapse, rise in crime, and so on. Whatever construction jobs exist, there will be a rush of hopeful workers competing for those jobs.
The "low" cost version of it is 20$ per query, which should in fact make it very viable for a lot of tasks. Yes, it scores lower on reasoning, but if it can do a day's worth of coding in more down-to-earth use cases for 20$... it's an absolute bargain. However, it would also depend on the max input tokens value. If it's too low, it might still be less useful for real projects than simpler models like Gemini which have this aspect covered better (1m+ input). Also, let's not forget that they hardly had time to optimize it. It could be significantly cheaper by the time of release.
Dont forget that we dont really know what "1 task" is. And (as you already said) we dont know how many input tokens it can handle before halucinating. And Gemini starts to produce giberish when feeding it with more that 32k input tokens of code trusting what I read in the web (havent tested it myself yet). We will have to wait until the model releases to try out how awesome or not it really is ^^
6:40 You have to factor in CO2 emissions cosrs too.. most big companies are buying carbon credits. Hiring humans is better for the environment and for the time being much less costly
To be fair, most software written by Humans is utter trash and completly unmaintainable and bloated spaghetti code. So the bar isn't very high. This biggest downfall of Humanity will be its hubris.
Its over for some programmers for sure. I work with a team that writes Azure functions for data integrations between different systems and they are worried. And they should be. That type of coding doesn't require much creativity, so yeah......AI can do that stuff easily. They will either be gone or they will change to more of an analyst role where they write specs then feed them to the AI.
Are people not seeing what I am seeing? Yes, they are getting great and writing human readable programming languages, but why? These languages exist to make it easier for humans to essentially write machine language. Why aren't the models writing direct machine code and bypassing all the text parsing? Or the AI develop their own optimized token language that makes sense to them? I just don't understand the point at making AI good at writing Python or C++? They don't need to do that.
Are you going to blindly execute code you absolutely cannot read, outside of a sandbox environment? Especially for critical applications? Besides, Python vastly outnumbers machine language in the training data.
If I interview a person, I have a few leetcode questions i know well that I would ask the candidate to solve. This is to get an idea of how the person thinks and how they work. It is a small qualifier of what they can do and how they work. If I were to evaluate an AI, it would be way more constrained. I would have it solve a large problem, like give a set of files, parse them and create a database model, and classify the data for me with specified categories. Then i want it to ask me about requirements, like a human would have to. Otherwise, it will all come down to me to do all the useful work.
What i don't understand is y not use it as augmentation instead of fully b dependent on it. Hello does anyone understand wtf I am saying? But if I am I'm wrong, I'm curious 🤔
Currently its not practical to have this replace programmers. Even if they scale up to an AGI its going to burn a lot of money for simple conversations. They must make it to atleast an ASI to solve real world problems that humans can't do.
Coding is less than 10% of a programmers job. About 1% when it gets to engineering. Generative models are already pretty good at coding, saving us a bunch of time. I'm generally excited for this to get to the next level. Can't wait to have more time to spend on meaningful problems
Architects and designers will not be replaced easily. But people who type in data into excel and stuff will be replaced soon. Data analysts or admins will be the next.
It does is by bruteforcing many responses from previous models, so for now it costs fortune and very slow. Second maybe less issue, because it has 24x7 for use
Only a few years into the AI frenzy and the leap in capabilities is very impressive. Hardware will eventually catch up, and these tasks will become cheaper over time. If this keeps pace, will have super AGI before we know it.
Stop saying we, you'll never use it and companies won't care about you if we get to this point. Start thinking rationally, if ai gets too smart then you're better off starving than wasting earth's resources for the millionaire class
But even with all of those improvements, companies value their privacy and data more than anything, so from my perspective it’s difficult to think that AI will replace programmers if the cost of creating software by an AI is sharing all of the data with another company. What do you all think?
Privacy doesn't matter if you start losing to competition which decided to use it, it takes one company to switch for others to follow. Just like how there was talk that artists need jobs too, ai wont take jobs. Well lookie here, ai used everywhere on netflix, people not hiring fleelancers for art work, ai used in games, ai used in r34, ai used in basically everything you can stuff it in. The world is fucked there's no sugar coating it, and companies would throw you out and switch to ai if it was more performant and less costly than you. I would suggest saving up money for now and praying that somehow ai gets either too expensive to run or complications arise. Good news is that if ai gets better than SE then it gets better than everyone but its just a matter of time. Other good news is that AI is not creative and is easily lost in more complex scenarios like balancing in games or overall delicate maths that you need optimised, it has absolutly 0% success rate. If you need an original product then it will have problems with that, but it sure can help with easy tasks,remembering code structure and overall fixing errors (not always, sometimes it just breaks your code for no reason).I would say personally that this is another fluke and we should just continue on. If something groundbreaking releases then we will have to make a switch but for now quitting because openai says possible bs to investors is a little too farfetched
Also, not only will software devs become unemployed but we wont need software (as we now know it) at all anyway. 99% of software written today is just UI's on top of business logic and databases. None of that will be needed in the short term.
there will always people that orchestrate and structure the code where AI fails. this accounts especially for bigger projects and more specialized needs (for example software machine interaction)
03 is as slow and expensive as it will ever be; later distillations will run faster and cheaper. At $1,000 per coding-task, it won't replace coders. Yet, at $100 per task, then it can do 2,000 tasks for the price of a seasoned coder's year of salary... WITHOUT having to wait for an entire year to get the answers! That elimination of LAG will add way more value, by giving the adopter more dynamism in the market.
Conservative estimate: O3 outperforms 99.99942% of developers worldwide (upper limit: 99.99965%)! Incredible performance, but definitely comes with a hefty price tag.
Cute how people are refusing to be seriously worried over their job when they should be worried about their existence. But that's a risky proposition to push status-wise, an idea too far out there for most people to entertain it without fear of seeming silly in the eyes of their audience. Normies gonna norm until it's way too late to even try to do anything meaningful about it. Revealed preferences for most people shows personal reputation among my peers >>> survival of my species.
no matter how smart ai gets you will always need a person to guide it to where you want to go, pure ai output no matter how smart is always going to be super generalized so just relying on that is jus ridiculous, ai just makes things a bit quicker and and lets you do things faster then usual
AI will not only reduce the need for programmers but will also decrease the demand for applications. Users will be able to tell computers exactly what they want, and AI will generate custom UIs on the fly, tailored to their needs. No need to learn complicated , generalised UIs. AI will completely transform how people interact with computers.
your analysis is incorrect, when companies can produce things at a cheaper cost, prices should drop, and you should get more free time because you can buy more for less work increased efficiency should lead to earlier retirement age, but it doesnt because of a rigged economy with artificial inflation
You will always need someone to sit in the middle and knot the strings. AI beats developers in constrainless framed logical problems but i dont think it will be good enough to migrate a legacy system architecture piece by piece while maintaining the outer interfaces and communicate with other companies to make sure everything runs fine. You can get help to find good pieces, you can get inspired or improve
Nowadays there are so many cheap "programmers", who have about the same skill in programming, as the next cat walking on a keyboard, so even if it were true, being better than 99% of programmers does not mean much nowadays. By official statistics the amount of programmers doubles every year - so 50% of the programmers have less than a single year experience. And 75% have less than two years of experience. And 87,5% have less than three years of experience. Thus, saying an AI has more skill than 99% just means it can beat some of the programmers who have 3 or more years of experience. Having started with 16 and being in this job for over 20 years now, I cannot say I am impressed by this "skill" of the AI.
relax dude whoever is saying jobs are being outsourced to India.. Its a lie, like US in India also there are no coding jobs. Indian managers never hire programmers who are over 4-5 years, people with 3 years experience are being recruited for Team Lead roles. Do you imagine?. Lot of changed over last 5 years. Managers will say why would we hire you when junior also can write the same code that you have written. I have faced this question many times during interview.
I love how people keep jumping on this AI replacing programmers thing. I use AI every day and I can tell you AI cannot code worth ship. Sure it can do basic coding but it cannot create a simple POJO that performs a complex function. It also cannot, in any shape or form, create code from pseudocode. I have tried a ton of different models and nothing. In fact the only way AI has worked for me is to assist in creating very basic code, and if supervised closely it can create some test code but I have little faith in that as well. I lost an entire day of coding because the previous programmer used AI to do 80% code completion and the code was so borked that the AI created test code only worked with the exact SAME input every single time. If you altered the input to the test code at all for the different values that can be absorbed by the code it failed. So when someone says AI can code better than a programmer... well that is just not true, it cannot take in complex prompts to create an application, no matter how detailed the prompt. It also cannot create code to log into certain access points, not wifi here but servers and DB's, it simply doesn't understand it. I even gave a local AI model my code that works 100% and it could not create a simple login to a DB to create a DB query that worked. I had to go in and tell it exactly what was wrong and after it fixed the code it broke other code. So no, again, AI cannot, at all, in any way fashion or form create complex, code and therefor it cannot replace programmers.
I wouldn't sound so absolute. AI is currently in its infancy. ChatGPT came out in 2022 and couldn't code at all. It's now 2024/2025... and there are claims that the unreleased model can code better than 99% of coders. Though that yet to be seen... if the claims are true, what level do you think it will be at a few years from now? Of course, no one wants AI to put humans out of work, but don't be so naive....
@@JayC216 the statement is oversimplified. Codeforce is not all programmers doing what they are paid to do. Writing code is just one part of that. Not to mention the model was trained on the dataset.
@@drwhitewash It learned Codeforce today... and it'll learn something new tomorrow 🤷♂... and with the introduction of quantum computing and LLM's abilities to now think, reason, and create original concepts, not just pull information from a database... I think we're gonna be in for a big surprise in the next year or so. People are looking at AI's current limitations and dismissing it as if the advancement of AI hasn't been extremely rapid... and starting to pick up speed. That's like looking at a three-year old and saying "Ha! He'll never be better than me at anything!" as if that child won't learn and grow.
@@JayC216 to learn something new tomorrow, it will have to be trained on that. Public training set from these benchmarks was specifically included in training data. Training of these huge models is expensive and will continue to be for the foreseeable future. I don't know if it's starting to pick up speed or it has already started and had a few hiccups on the way already. Openai has its struggles and delays, that means the low hanging fruit is starting to be rare. Don't forget the sigmoid curve.
However, today is not tomorrow. If things aren't on a trajectory but suddenly freeze and remain the same, then you have a valid point. However, your point is invalid.
Haha, when I saw thumbnail, I thought "who is this muscular guy in a sexy white shirt? Let's click the video" and you got me. Good job getting my monkey-brain to pay attention haha :) On-topic: hearingyour analogy about farming technology reducing the need for human farmers but this being good for the world, it makes me think about how scary change and unpredictability is. I can't help but think that if we, as a society, had more of a safety net such that people could adapt to change in our job landscape more easily, then there would be less anxiety.
eh I think it'll just make coders more productive, thus increasing their marginal revenue product, and making them more valuable. The ability to google stack overflow, and really good IDEs was a slightly bigger change but this is still a big deal.
Programming as a career becomes more select. You can stay in denial but keep in mind this is the worse it will ever be! Microsoft CEO was right when he said that AI will replace the application layer to interact with data.
@@chillgamervids dbms is still needed, even more so :) Additionally to everything else, you also need to store vector embeddings for the new kind of LLM based applications.
It's the end. No ostrich head syndrome please. Like calculator ended manual calculation, and many other jobs disappeared, it's for software engineering now
As a Software Engineer with 20+ years in the industry, working in both startups and big tech, I would love to see the face of one of these non-technical AI extremists when they realize programming is just a small part of building, operating and maintaining a successful software product (and it's usually the easiest part).
I have 14 years of experience and can confirm. All of a sudden everyone is an expert in AI but time will tell them
What do you think is the number one job OpenAI would benefit from automating? What skill can AI acquire that gives OpenAI an advantage over its competitors? What one skill would give OpenAI’s competitors an advantage over them? What type of expertise do you think is most abundant within OpenAI as a resource to fine tune what they are creating? When a new model comes out, what is the first thing they benchmark it on? What real world thing/tool is an AI best positioned to replace?
So is it worth to study softwarw engineer/ Systems engineer?
@@Mateobjs_ Yes. You are more than a programmer. You are a problem solver. Building and maintaining software will continue to change and become higher level. We were using punch cards in the 70s, then came higher level languages, now we have AI assistants to greatly improve productivity (which I think is a good thing), and it will be different in 10 years from now.
@@Mateobjs_ Yes. You are more than a programmer. You are a problem solver. Building and maintaining software will continue changing to become higher level and you have to adapt. We were using punch cards in the 70s, then came higher level languages like C, now we have AI assistants to boost productivity (which I think is a good thing), and it will be different in 10 years from now.
This is the 41 month of AI replacing every developer next month
Well, that's because the previous releases we're all hype that didn't meet anyone's expectations. But having an AI that can code better than 99% of humans on the planet is a different story. Will it happen next month? Nah... But if it's at this level now, where will it be in five years? It'll be at a level we can't comprehend.
@@JayC216 but can it? It was trained on that benchmark, on assignments similar to those in the private test set.
@@drwhitewashcompetitive programming problems follow a pattern there are only like a dozen kind of problems like dp ,graphs ,greedy the programmers are trained to find the pattern of the problem then the problem can be done easily
@@JayC216 spend probably 50 hours on learning to program computers and ur in the 1% too
@@ceciljoel9577 exactly. It is good to know graph algorithms, greedy algorithms etc, but real world problems and their solutions are much more than just these things.
Although they are definitely usable in also solving real tasks, knowing these patterns is not enough.
enterprise-ai AI fixes this. AI beats 99% of programmers.
Beats 99% in benchmark and still useless in real life problems. Stop training AI for beating benchmark, start training for everyday use
how do you know it's useless for real life problems? have you used o3?
You REALLY aren't tuned into what's going.
I build AI automations for clients...real life use cases.
This will translate and AMPLIFY everything.
Sure if your job is eating cheeseburgers all day, you wont be affected.
But for those of us gainfully employed, this will be big.
SWE bench
@@AirSandFirehe is just mad and afraid to lose his job white he is rna poisonned and depopulated anyway
@@AirSandFire it costs 3000 dolars for easy benchmark, it is useless
A calculator is better than a human at certain aspects. It doesn't mean we don't need mathematicians
The main point of all these AI advancements is not about tools that assist humans. They are speeding up, which replaces humans. I hope you got my point.
The job "computer" was coined in the 1600's and referred to a human profession a person who made calculations. No, we don't need human computers or human calculators. We once did, today we do not. I'm sick and tired of all this cope and the absurd linguistic arguments people make to cope. Change is happening. It is significant. Define it however you need, jobs will be lost.
Until it can read my entire codebase, analyze it, make it run more efficiently, fully compile it, and, if it fails to compile, search the code for errors, fix them, and get it to run, it's still a few years away.
Thats sound like a monolithic application . Like trying to use ink pen in space Not going to happen nor anyone have time to do it. Simple use pencil in space or microservices for same service
dude, it can. Right now. Just that 03 hasn't been released to the public yet.
@@echabbewalwell once you get to hundreds of microservices you are glad to have a monolith.
@@christophercarson3675can it, really?
@@christophercarson3675in a highly theoretical case where it's only trained on specific test questions it can. That doesn't change the fact it's only infrastructure most of the tech sector can't afford nor would they have the expertise to implement in house training data without hiring a dedicated team to engineer solutions for a specific use case. It isn't as simple as just turn on and go
As a developer with some experience, I work closely with a senior and near-senior developer, focusing more on planning and some coding in the private sector. Satya Nadella put it best: don’t worry about programming disappearing-worry about software as a whole becoming obsolete. At its core, software is just a UI for a database with business logic.
With Moore’s Law driving hardware efficiency and AI advancing rapidly, traditional software is evolving into AI agents and LLMs, eliminating the need for apps or typing into interfaces. Businesses, however, often resist change due to inertia-a phenomenon akin to the "frog in boiling water" analogy, where gradual changes go unnoticed until it's too late. This shift will redefine many professions, including programming, but it also opens up opportunities for those who adapt to new technologies and embrace AI-driven innovation.
Factory automation moved slower than was possible because there were sunk costs in existing factories and equipment and new equipment was only added as old equipment and processes started to need replacement. I don’t see that being a factor with AI since it allows a business to reduce facility size, close offices, and decrease head count.
Let's not compare agents with algorithms, please.
Agents were designed to automate natural text processing. Like you got a tweet, you first disect it into a certain category you define. Then you realize whether it's positive or negative. If negative, you try to find out what the customer was unhappy about (again a classification task). Then use some RAG to determine additional offers for the customer and propose them.
See? Natural language processing/understanding.
You should definitely not use agents to do any deterministic, algorithmic task. That's what the software is for, even if it's just a GUI on top of your expensive Oracle.
Just some bits (no pun intended) of software that don’t fall within the above confines:
System Software - Operating Systems (Windows, Linux, macOS), Kernels (Linux Kernel, XNU), Device Drivers (NVIDIA GPU drivers, Realtek audio drivers), Firmware (BIOS, UEFI), Virtual Machines & Hypervisors (VMware ESXi, VirtualBox)
Development Tools - Compilers (GCC, LLVM), Interpreters (Python, Node.js), Debuggers (GDB, WinDbg), Integrated Development Environments (Visual Studio, IntelliJ IDEA), Build Systems (Make, CMake)
Networking Software - Protocol Implementations (OpenSSL, OpenVPN), Web Servers (Apache, Nginx), DNS Servers (BIND, PowerDNS), Packet Analyzers (Wireshark, tcpdump), Routing Software (Quagga, FRRouting)
Entertainment Software - Games (Minecraft, Fortnite), Game Engines (Unity, Unreal Engine), Media Players (VLC, Windows Media Player)
Scientific and Simulation Software - Physics Simulators (COMSOL, ANSYS), Biological Simulators (GROMACS, BLAST), Medical Simulators (SimMan, OpenSim), Mathematical Tools (MATLAB, Mathematica)
Embedded and Real-Time Systems - Embedded Software (Arduino IDE, FreeRTOS), Real-Time Operating Systems (VxWorks, QNX)
Artificial Intelligence and Machine Learning - AI Frameworks (TensorFlow, PyTorch), Natural Language Processing Tools (spaCy, NLTK), Computer Vision Software (OpenCV, YOLO)
Web and Mobile Applications - Web Frameworks (Django, React), Mobile Platforms (Android Studio, Flutter), APIs and Middleware (Express.js, GraphQL)
Security Software - Antivirus Programs (Norton, Bitdefender), Encryption Tools (VeraCrypt, OpenSSL), Firewalls (pfSense, IPFire)
Control Systems and Industrial Software - SCADA Systems (Ignition, WinCC), CAD Software (AutoCAD, SolidWorks), Robotics Software (ROS, V-REP)
Multimedia and Design Tools - Image Editing Software (Photoshop, GIMP), Video Editing Software (Adobe Premiere, DaVinci Resolve), 3D Modeling Software (Blender, Maya)
Cloud and Distributed Systems - Containerization Tools (Docker, Kubernetes), Distributed File Systems (HDFS, Ceph), Serverless Platforms (AWS Lambda, Azure Functions)
Testing and Monitoring Tools - Performance Testing (JMeter, LoadRunner), Monitoring Tools (Prometheus, Grafana), Automated Testing Frameworks (Selenium, TestNG)
Specialized Software - Blockchain Platforms (Ethereum, Hyperledger), Geographic Information Systems (ArcGIS, QGIS), Quantum Computing Software (Qiskit, Cirq)
there are more fields than business for software
I doubt that it will happen before AI can fully automate programming, project managing, and well, ALL jobs, at e level of 99 percentile, or bigger (which I think is not far actually). Big part of software actually does computationally small amount of work to make majority of user operations, like sending a message, creating account, and etc, and still needs a lot of optimization for being able to process it all. For everyday app usage consistent enough AI agents costs many orders of magnitude more than it needs to for this scenario. AI will definitely appear in any big app, for helping, or maybe performing difficult and complex operation, but it won't make software obsolete, and it is actually a very small problem, compared to AI just replacing programmers, which is also a very small problem, compared to someone creating an unaligned self sustainable, exponentially growing AI agent swarm, or super intelligent AI, using our social fragmentation, persuasion, and humanity flaws for escaping gatekeepers.
The seller explaining us how good is the product he sells.
Exactly. These people would believe every word a used car dealer tells them. Sometimes I believe the man in the street has more common sense than these people.
The craziest part is programmers who said this, like they have something to gain about all this. Presumably smart people who work in IT riding the hype train is insane.
Probably a better analogy of what is coming is what happened in manufacturing. I was working in industry in the 80’s when computers and automation first started entering the factory. It wasn’t very good, expensive, and required a lot of human intervention. 30 years later I do the work of 5 machinists better and faster. But the problem AI poses for human tech workers is it’s happening basically overnight. The 4 other machinists had 30 years to retire, retrain, or just grow with the technology. There is no way AI will create more jobs than it destroys. We all can’t just sell coffee to each other.
I find it so wild. I swear the people who say it's going to replace programmers are the people that never knew how to program to begin with.
On the contrary, People saying it won't replace programmers are like people saying humans will never invent aircraft and fly.
@@ShaneMcGrath. Using this logic you can more or less prove anything.
It already happened with older AI models. The new models will soon get to the point that using a human programmer will be unsafe and undesirable not just expensive.
Bro you don't even code 😂. Nothing happened with the older models I use gpt-4o and it's still extremely crappy
“All truth passes through three stages. First, it is ridiculed. Second, it is violently opposed. Third, it is accepted as being self-evident.” - Arthur Schopenhauer
OpenAI is one of the worst in terms of their sh`tty marketing. They've been overhyping releases for years. As a software engineer, I do use AI every day, in fact a lot. It does speed up the development process if you give it simple tasks to solve. It's currently on the level of a junior developer, more or less. But there's a whole world of tasks where it starts hallucinating and writing BS - mostly architectural decisions, refactoring stuff, various integrations, etc. I don't think O3 will be such a huge game changer. OpenAI can keep saying big words but synthetic AI tests are NOT the real-world tasks. And the original SWE-benchmark, where it kinda sucked, just proves this fact. I'm not in denial, but I think the truth is just between polar opinions. Is AI totally useless? ofc not. Will it completely replace all programmers? No, not in the next 5-10 years for sure. The correct answer "yes, it might take some tasks and may or may not shrink the market and cause some decrease in IT salaries, I'd say anywhere from 10% to 50%"
Don't get me wrong, I'm all for new tech and A.I. and a huge tech nerd myself, But let's face it for normal working class blue collar plebs like me which make up majority of workforce, Most of this A.I. is completely useless in real world still!
They are just doing the typical sales pitch to pump up hype to get more investment money, Wake me when it's merged with a humanoid robot that can take the trash out, Mow the lawn and wash the dishes all when I ask it to in a just a few words.
Soon but at what cost? Financialy and ecologiqly?
I use AI for most of my issues and questions in life becuase AI is smarter and more reasonable than 99% of the cooked society we have to deal with. Like once you get used to AI dealing with people is fucking absolutely useless.
@@T500Kz
ahahahaha!!!
Soon the sex IA also !
the thing is... when errors arise, who will be to blame? i think companies need someone to blame for errors, first and foremost, and it's not possible to blame a machine that doesn't care. that is, if users even care about the quality of the software they use. the direction seems to be software becomes increasingly buggy and bloated, no one to blame because it passes test, user doesn't care and will just refresh the page. be the change you wanna see, don't use bugged paid software, and go the extra mile to ensure quality in the software you produce
It´s funny that OpenAI already promotes O3 as "AGI (Artificial General Intelligence) is finally here" when in reality it´s just the next step of AI.
It's because once they say they have achieved AGI they get a huge chunk of money as part of a milestone payment that is defined in one of their contracts with Microsoft
OpenAI has specifically said they are on the path to AGI, which is not controversial. All frontier labs are on this path… whether it’s a 2 year path or a 20 year path. But OpenAI did not call o3 AGI. Many OpenAI fans and sensationalist content creators commenting on o3 are calling it that, but OpenAI is not.
They didn't call it AGI at any point? They even said in the announcement video that this was an important STEP TOWARDS AGI
openai has absolutely sh*tty marketing strategy, I simply stopped believing anything they been saying in the last years. The company itself is rotten to the core. And we all know the reason behind it
6 15 Hold on. Look at the second highest score. o3 got 76% and it looks like it cost, what, $20 per task? A human programmer (in the US) probably gets paid between $50 and $75 per hour. And the human also requires the company to pay matching contributions to 401K, some towards health insurance, holiday pay, vacation pay, etc. The AI seems pretty competitive, now.
When Orion releases next year by December 2025 and the cost to operate O3 falls below $1,000, things are going to get weird. The $1,000 price tag won't last for a year. OpenAI or some other company is going to bring the cost of an O3 model down to O1 prices.
@@CMDRScotty the price tag is even higher. Orion is not a reasoning model, so how does it relate to this?
@drwhitewash I am saying Orion combined with systems like O3 is going to cause unprecedented change to the AI industry. I also believe that Open AI and its competitors are going to push the cost of O3 systems to under $1000. Technology always gets cheaper just like computers laptops, and cell phones have done over the last 50 years AI is no different.
There will be no fixed $1000 USD price tag per task.
Re the highest scoring ARC benchmarks: it was more than $1000 per task. However, the point was here to demo the power of test time compute.
The pricing model will likely be based on the amount of compute you use. So you’ll be able to make it “think” shorter or longer, longer being more expensive.
Regarding common programming tasks: the price tag will be a small fraction of the mentioned $1000 per task, as there’s so much more code on the internet. And huge efforts go into optimising models for coding. The model o3 wasn’t wasn’t optimised for the ARC task - somehow these LLMs and systems that utilise LLMs (such as o3) don’t natively understand 2D grids well. Basically location or movement or symmetry. Presumably something to do with lack of training data and how the tokeniser operates.
They trained it on 75% of the questions from that test. Read the white paper.
Open AI gets a ‘C’ on an evaluation it was tuned for.
Less scary now?
yes, they trained it on the training set...
you cannot "test" on Frontier Math evaluation : even humans mostly fails at these
I highly doubt that they discarded cross validation and trained on the test data.
The arc test are novel questions that haven't been seen, I believe they are generated but im not sure
@@micahwilliams1826 no , they're tests that humans can do very easily , but that AI cannot .
it's also the reason why o3 is not AGI, there are tests you can easily understand, yet o3 struggles.
You can submit your own tests to ARC , I think they have a github for this
Who are those smart people saying this is the end of programming? I just want to know.
People who could never get a programming job
People that made the A.I., Much more complex program than all the fools pretending their job is safe at the office.
Even the A.I. coders will be out of a job in coming years. lol
@@ShaneMcGrath. There is not much coding in AI at all. Many training scripts are only a couple hundred lines long. The research is the much more challenging part.
NVidia’s CEO hinted at this outcome a while ago.
Hinton, Geoffrey
Definitely not the end of programming, Some things can only be described through programming jargon that can only be understood if you know how to program. How are you supposed to give specific instructions without knowing how to code? If you want to do anything original that is finely tuned to your requests, a hands on approach will likely be required unless the AI can read your mind.
It's heartwarming to see people who have for over a year declared that AI will never be competent at programming, now similarly declare that AI will never accomplish coding related tasks.
I wonder what field they'll next declare AI will never accomplish while they're permanently unemployed with no hope of any income.
It's still pretty bad at coding in real life .Have you used it?
@@Justin-wj4yc They have significant limitations and make mistakes. I don't know if you're just trying to make conversation or if you honestly think this observation somehow invalidates my point.
@@kyneticist What was your point in your comment? It's like you're salty developers make more than you. AI hasn't improved much at coding in real life since ChatGPT launched. When AI is good at coding in real life not just benchmarks it will be good at everything
Why are you so salty... The point when AI actually becomes better than humans in coding is where the next big modell will be made by AI. And here is also the point where it can imporve itself over and over until all jobs that are "mind-based" are going to get replaced, exept for psychologists. (Most of "them" are employed making a lot of money btw, not sure why they would be unemployed)
AI will remove the boring routine part from coding like coding itself. Humans will be left with more challenging tasks like making coffee.
These bloody Stochaistic Parrots keep getting better and better. Really annoying.
For the unemployed with ideas of creating companies this is the greatest era ever. I no longer need tutorials on UA-cam but still helpful to know the basics and lot more otherwise you won't know what to ask the ai to make you. Example of someone who knows vs someone learning and how can they can use it.... I can then ask the ai to write me a basic react components from navbars to help setting up the whole folder structure and then also and such then also add css . From there I customize based on the design I make. Today I can build full stack websites with the help of ai. Before I had to hire a programmer.... I guess you are right their days are numbered but for saas the jobs will always be there for engineers who have positions like system administrator and such positions that can't be done by ai only their agents which would need to physically be there or windows 11 will have to find a way for an ai agent to do all the administrative work with the help of 1 person
You need API credits! :)
AI could never *design* systems, it can only code. Learning just coding was not enough either way. That’s why bootcamps and self-learning have low success ratio. What will change in the near future? Monkey programmers will become obsolete. Developers and software engineers that participate actively in designing the software architecture have no fear. Instead of whining about AI taking your job, upgrade yourselves from a programming monkey to a designer.
And you are talking about the 10% of jobs
@@mo_mo_meaw I don't believe it's such a low percentage. But I don't necessarily believe it's a bad thing. As Excel rose the number of accountants, AI could rose the number of software engineers. The job market, yes, it's changing and we must conform to it, but skillfulness is NEVER wasted.
Famous last words… this is the dumbest it will ever be And it’s performing like this. You will be replaced, it’s just a matter of when.
La ia si podrá diseñar sistemas, solo les falta algo de tiempo
@@jonathanduran3442 We can't even solve the P=NP problem yet. LLMs are just fancy autocomplete engines.
Older chatgpt models were very costly to run as well before being at commercial stage, optimisation will happen to bring prices down 10x-100x soon. They will realise eventually the agents vision, with multiple specialistic agents trained to become architects, testers, managers and collaborate all day long into software solutions. It's not here yet, but to me it's 2-3 years away to be truly job nightmare.
If ai can beat programmers, wouldn't it also beat every white collar worker?
Why would i need OpenAI or Microsoft when AI beats all programmers?
Yes
Pretty much, thats the real change but there's a lot of people in those places that can choose not to use it; unlike a developer who was already seen as a sunk cost.
People who work with Excel, data input workers will be replaced soon.
@@junggyujoo Only if they dont care about the accuracy of that input....
I am sure that AI writes hello world a lot faster than human beings.
2028: "okay we have nerfed o4 to death BUT oh boy o5 will be AGI "
AI is underrated
@Warley.Araujo if you're coding at script kiddie level
@@hendrx well it certainly got better
@@hendrx well, another ignorant one that will cry in a year or two...
edit: Also, o1 was revealed a couple of months ago, and they skipped o2 entirely. We will have o6 or o7 before the end of 2025 at this pace, and it will definitely be better than every human developer.
o5 is more like late 2025
the test isn't whether it can beat programmers ; the test is can a none programmer use it to create an app to client specifications. The day a client can say to a none technical person "use your AI to create me an app to do these precise tasks and integrate it into my website" ... thats the day.. thats end of skilled dev.. but , as always I just predcit the skills will change but remain technical and difficult. Do we program in binary anymore? Majority of us not ; we use words, numbers and symbols. I think what we're seeing is programming language becoming the spoken word instead.. but that wont make it easy enough to de-skill it.
Occupations change, come and go. What the work consists of changes.
Before refrigerators there were iceboxes, cabinets cooled by ice. At that time there was a market for ice delivery. That market is gone now.
What developers do has changed some during the 40 years I have been working with computers. The web didn't exist in 1980. Programming languages that were common back then are not so common today; instead new languages have come. Developer environments have improved a lot.
The developer job will continue to change. AI will help developers even more than today. The role of the developers might change to be more about prompt engineering. But for quite a while, I think, classical developers will still be needed, especially for those parts that AI's don't know very well.
When things break, it will be bad. Most of the time, developers spend on design, architect, integrating, testing, sustaining, communicating, and troubleshooting. Programming is only good as who cross check if it works and why it failed. If a person comes in with little programming background and develops the product. Within a few years, the complexity of maintaining that software will be tremendous. And remember, the software or AI companies will try their best to sell their product to lock in customers. But when applying to actual business cases, a lot of time, they might not meet customers' desires. Think of any software companies and their products, we want to standardize but always end up customizing the product to meet business needs. Just my 2 cents, I am also eager to see how it goes from here.
These are no longer "models". The word model has been co-opted. Instead of the model directly (in a single pass) producing an output, they are running their own agentic workflows in the backend to correct and compensate for the actual models poor performance. Hence the rising cost and long wait times. It's basically a scam. They are creating the illusion of progress by hard coding guardrails and validation environments to assess outputs before feeding them back to re-prompt the underlying model to hopefully get the right answer. These are effectively error grinding while loops.
Any developer whose job was taken by AI had no business being in tech in the first place. Anyone who believes AI is driving the current job market is a mouth breather and developers who use ai code editors to generate code they don't understand have no inner monologue and don't know how they would feel if they didn't eat breakfast yesterday.
Dude you're not thinking! You're actually questioning programmer jobs and costs??? This just started! And u can already begin to see the exponentials starting everywhere. Those costs are gonna get cheaper exponentially as has been happening and YES PROGRAMNERS ARE DONE! Pretty much everybody 90% anyway is done 10 years out. In 3 years we've garnered probably 100+ years of knowledge already. We are 10xing every 3-5 months now n very very beginning n tip of the iceberg smh. People gotta wake up
I agree
What people don't understand is that when it can make software then it can also use it. So everyone who uses Software is now useless.
😂😂🤣🤣Too many blue pills for you?
@@wolfgangpreier9160 you think a programmer who builds a software couldn't use it more effective than the guys who do it now? programmers will just take their jobs because they are higher qualified.
@@wolfgangpreier9160 the only reason why they didn't in the past was because building the software was paid 20 times or more higher.
@@wolfgangpreier9160 coincidently I am a developer working with that ai stuff. I could show you how a script learns how to use a software just by itself
@@jojojojojojojo2 Why should a programmer build a Software and then use it himself?
I have done it to calculate my optimal PV and Battery size but that was for my house.
When i have programmed in the past - i have stopped doing that about 5 years ago because not enough profit, it was always for a customer.
Just writing code is one thing but when it can talk to ones who ordered the code to be fixed/created, get right all the requirements and then realize some of them do not make sense, go back and talk to ordered, realize they cannot agree to anything and force a meeting with all involved parties, then make them agree on new requirements, make sure they are fine, all works and is tested and built...then a ton of jobs in AI will disappear.
Coding is...coding. It is not all what programmers do. Sure many times part of all that I described is done by business analysts, testers, product owners, scrum masters etc. but still a programmer is very much involved in all of that.
And that's for existing piece of software. Now do it all plus you have to do it completely from scratch and choose all the technologies and how you go about say creating a new ledger system for a bank.
I'm kind of worried i just started to learn program and i like it but now idk anymore :/ should i stop or continue?
If you like it continue. You can still write software for fun. You will not be able to compete with AI.
As a fellow programmer myself, I've come to the conclusion that pure SWE has no future. I think it's best to veer to robotics and automation at this point. Anything that has to do with processes in the physical world cannot be automated nearly as quickly as progromming due to a lack of training data.
don't get yourself worried. the current performance of AI in coding only looks good from the perspective of Dunning-Krugers "valley of ignorance". programming consists of so much more than typing out the code. it's about understanding the problem you are trying to solve, the constraints and options you have and then tying everything together in a way that "works".
unless it is your passion, stop, there will be layoffs and less jobs
@chillgamer,
Yeah, real programmers don’t do any coding. That’s just for juniors. Real programmers do grand architecture and difficult business decisions.
So, is our software engineering career dead by AI or what, please advice...🙏🙏🙏
no, economy fucked the career in the first place, nobody that actually knows shit about llm says they are replacing softwares engineers, it's probably the end of frontends taking one week to turn a button from red to blue tho
yes
Not yet atleast, we dont know how good it can code inside large project, even the SWE-bench test is not really clear on how much of its result are inside the training data. And these projects are all Python projects. How would it perform on c++ or rust projects where there is less training data(rust) and manuall memory managment (C++).
And one pull request costing ~3k is not viable, but this will clearly decrease in the future.
You should wait until o3 comes out and real testing is done
O3 is good but takes 1000$ per prompt and is its context window (how much text it can handle) is not big enough to think over an entire programming project. As programming is very intensive and demands creativity, i dont think prompting it over and over and over again is a viable option, so no nothing changes for now atleast for most companies, can't say the same about giants like google who could use it from time to time without feeling it
I believe that at least for the next years, we will still need Developers reviewing and supervising the work done by AI to make sure it is going in the right direction. But this will mean, we don't need a lot of junior devs, but instead of a team of 10 devs, we will need only one senior or software architect using AI to do the work.
May I inform you that there will be highly skill AI agents and computer use agents to collaborate with the o series models that code. This means an orchestrated AI swarm will be able to create the idea, code it, test it, send feedback, learn and revise all on its own in "2025".
I agree with you. The current systems are significantly lagging humans with general vision abilities, but this gap will also be closed soon.
Massive context windows with nearly perfect recall are also just around the corner.
This thing is moving at lightning speed. Altman had indicated that the o-series will be updates every few months or so. He’s said “we’ve got the next 2 years in the bag” or something like this. That takes us to o-10…
@@Justashortcomment Programmers need to look into being "Manager Engineers".... meaning how to manage these AI agents, requesting prompts, review board and how to comment on AI agents feedback of their programing skills, etc. If it is not the desired outcome of your project, do you blame the AI agent or your own prompting skills? Managing AI agents for coding will be the new skillset I feel.
i get paid $500 a month soo i'm cheaper than AI
wait until chinese AI come
@@thirien59 when that time comes, i'll do construction work ))
@@aziz9488 The construction work that autonomous robots will be able to do day and night 10x faster?
@@aziz9488Construction work? When just thirty to forty percent of cognitive workers are unemployed, how much construction work do you think there will be? The economy won’t be booming, that’s for certain. When that many people lose their jobs, we’re talking millions of people globally defaulting on mortgages, credit cards, car loans, etc. leading to a banking crisis, massive drops in tax revenue , a consumption collapse, rise in crime, and so on. Whatever construction jobs exist, there will be a rush of hopeful workers competing for those jobs.
@@aziz9488Wait until Chinese robots come
The "low" cost version of it is 20$ per query, which should in fact make it very viable for a lot of tasks. Yes, it scores lower on reasoning, but if it can do a day's worth of coding in more down-to-earth use cases for 20$... it's an absolute bargain. However, it would also depend on the max input tokens value. If it's too low, it might still be less useful for real projects than simpler models like Gemini which have this aspect covered better (1m+ input). Also, let's not forget that they hardly had time to optimize it. It could be significantly cheaper by the time of release.
Dont forget that we dont really know what "1 task" is. And (as you already said) we dont know how many input tokens it can handle before halucinating. And Gemini starts to produce giberish when feeding it with more that 32k input tokens of code trusting what I read in the web (havent tested it myself yet).
We will have to wait until the model releases to try out how awesome or not it really is ^^
6:40 You have to factor in CO2 emissions cosrs too.. most big companies are buying carbon credits. Hiring humans is better for the environment and for the time being much less costly
To be fair, most software written by Humans is utter trash and completly unmaintainable and bloated spaghetti code. So the bar isn't very high.
This biggest downfall of Humanity will be its hubris.
Its over for some programmers for sure. I work with a team that writes Azure functions for data integrations between different systems and they are worried. And they should be. That type of coding doesn't require much creativity, so yeah......AI can do that stuff easily. They will either be gone or they will change to more of an analyst role where they write specs then feed them to the AI.
AI will replace the analyst role also and the designer role also.
Are people not seeing what I am seeing? Yes, they are getting great and writing human readable programming languages, but why? These languages exist to make it easier for humans to essentially write machine language. Why aren't the models writing direct machine code and bypassing all the text parsing? Or the AI develop their own optimized token language that makes sense to them? I just don't understand the point at making AI good at writing Python or C++? They don't need to do that.
Are you going to blindly execute code you absolutely cannot read, outside of a sandbox environment? Especially for critical applications?
Besides, Python vastly outnumbers machine language in the training data.
If I interview a person, I have a few leetcode questions i know well that I would ask the candidate to solve. This is to get an idea of how the person thinks and how they work. It is a small qualifier of what they can do and how they work. If I were to evaluate an AI, it would be way more constrained. I would have it solve a large problem, like give a set of files, parse them and create a database model, and classify the data for me with specified categories. Then i want it to ask me about requirements, like a human would have to. Otherwise, it will all come down to me to do all the useful work.
What happens next is you open your eyes wide and look down at the floor and snap a picture for your next thumbnail.
What i don't understand is y not use it as augmentation instead of fully b dependent on it. Hello does anyone understand wtf I am saying? But if I am I'm wrong, I'm curious 🤔
ill believe it when i see it build a full application, deploy it to production and generate revenue without human intervention
Currently its not practical to have this replace programmers. Even if they scale up to an AGI its going to burn a lot of money for simple conversations. They must make it to atleast an ASI to solve real world problems that humans can't do.
What was the O3 ai trying to create is the real question. Is coding a single function or is it coding an entire code base?
It’s trying to form a World Government.
where was it "launched" or "released" ???
Coding is less than 10% of a programmers job. About 1% when it gets to engineering. Generative models are already pretty good at coding, saving us a bunch of time. I'm generally excited for this to get to the next level. Can't wait to have more time to spend on meaningful problems
Man quitting my IT job last year to go back treeplanting is starting to shine.
Architects and designers will not be replaced easily. But people who type in data into excel and stuff will be replaced soon. Data analysts or admins will be the next.
It does is by bruteforcing many responses from previous models, so for now it costs fortune and very slow.
Second maybe less issue, because it has 24x7 for use
Only a few years into the AI frenzy and the leap in capabilities is very impressive. Hardware will eventually catch up, and these tasks will become cheaper over time. If this keeps pace, will have super AGI before we know it.
Stop saying we, you'll never use it and companies won't care about you if we get to this point. Start thinking rationally, if ai gets too smart then you're better off starving than wasting earth's resources for the millionaire class
I dont think it can replace people completely it may be better in some ways but needs human direction.
oh no the LLM can solve a very specific code problem in a very specific environment unlike anything you'd see in the working world. We're safe guys
that 87% cost $350,000 worth of compute
POV: Tell me you're a junior developer without telling you're a junior developer... :D
But even with all of those improvements, companies value their privacy and data more than anything, so from my perspective it’s difficult to think that AI will replace programmers if the cost of creating software by an AI is sharing all of the data with another company.
What do you all think?
Privacy doesn't matter if you start losing to competition which decided to use it, it takes one company to switch for others to follow. Just like how there was talk that artists need jobs too, ai wont take jobs. Well lookie here, ai used everywhere on netflix, people not hiring fleelancers for art work, ai used in games, ai used in r34, ai used in basically everything you can stuff it in. The world is fucked there's no sugar coating it, and companies would throw you out and switch to ai if it was more performant and less costly than you. I would suggest saving up money for now and praying that somehow ai gets either too expensive to run or complications arise.
Good news is that if ai gets better than SE then it gets better than everyone but its just a matter of time.
Other good news is that AI is not creative and is easily lost in more complex scenarios like balancing in games or overall delicate maths that you need optimised, it has absolutly 0% success rate. If you need an original product then it will have problems with that, but it sure can help with easy tasks,remembering code structure and overall fixing errors (not always, sometimes it just breaks your code for no reason).I would say personally that this is another fluke and we should just continue on. If something groundbreaking releases then we will have to make a switch but for now quitting because openai says possible bs to investors is a little too farfetched
Also, not only will software devs become unemployed but we wont need software (as we now know it) at all anyway. 99% of software written today is just UI's on top of business logic and databases. None of that will be needed in the short term.
While it's not the end of programmers for sure, it will offer massive productivity boost that many programmer will be let go anyway.
there will always people that orchestrate and structure the code where AI fails. this accounts especially for bigger projects and more specialized needs (for example software machine interaction)
People who say these things have very little understanding of what software engineers really do.
03 is as slow and expensive as it will ever be; later distillations will run faster and cheaper. At $1,000 per coding-task, it won't replace coders. Yet, at $100 per task, then it can do 2,000 tasks for the price of a seasoned coder's year of salary... WITHOUT having to wait for an entire year to get the answers! That elimination of LAG will add way more value, by giving the adopter more dynamism in the market.
Conservative estimate: O3 outperforms 99.99942% of developers worldwide (upper limit: 99.99965%)! Incredible performance, but definitely comes with a hefty price tag.
but it outperforms them in competitive programming... which is impressive. but is far away from enterprise programming.
In competitive programming, which has nothing to do with real world programming.
MS CEO says no need for software soon . Just give AI access to the data and tell it what you want.
Cute how people are refusing to be seriously worried over their job when they should be worried about their existence. But that's a risky proposition to push status-wise, an idea too far out there for most people to entertain it without fear of seeming silly in the eyes of their audience. Normies gonna norm until it's way too late to even try to do anything meaningful about it. Revealed preferences for most people shows personal reputation among my peers >>> survival of my species.
We should not be afraid of AI because it is natural that technology changes. Let us just accept and embrace this changes.
no matter how smart ai gets you will always need a person to guide it to where you want to go, pure ai output no matter how smart is always going to be super generalized so just relying on that is jus ridiculous, ai just makes things a bit quicker and and lets you do things faster then usual
AI will not only reduce the need for programmers but will also decrease the demand for applications. Users will be able to tell computers exactly what they want, and AI will generate custom UIs on the fly, tailored to their needs. No need to learn complicated , generalised UIs. AI will completely transform how people interact with computers.
And it is very expensive to run even one real prompt.
Thw problem that I have is that if we so not work to make money than it means that we don't have money to buy things.
your analysis is incorrect, when companies can produce things at a cheaper cost, prices should drop, and you should get more free time because you can buy more for less work
increased efficiency should lead to earlier retirement age, but it doesnt because of a rigged economy with artificial inflation
If you think it through, at the limit where AI can do everything at zero cost, it also pushed the cost of everything to 0
Yeah you put this model behind Cursor or Aider, it's gonna be fierce.
I expect collateral damage behind this.
You will always need someone to sit in the middle and knot the strings.
AI beats developers in constrainless framed logical problems but i dont think it will be good enough to migrate a legacy system architecture piece by piece while maintaining the outer interfaces and communicate with other companies to make sure everything runs fine.
You can get help to find good pieces, you can get inspired or improve
I hope it will finally be able to close brackets
I truly think IA will not replace only code, but entire systems. It'll be data + tuned IA, period. Time will tell.
you still need to be able to read the code and as a programmer. if you have no idea what the code does.
Nowadays there are so many cheap "programmers", who have about the same skill in programming, as the next cat walking on a keyboard, so even if it were true, being better than 99% of programmers does not mean much nowadays.
By official statistics the amount of programmers doubles every year - so 50% of the programmers have less than a single year experience. And 75% have less than two years of experience. And 87,5% have less than three years of experience.
Thus, saying an AI has more skill than 99% just means it can beat some of the programmers who have 3 or more years of experience. Having started with 16 and being in this job for over 20 years now, I cannot say I am impressed by this "skill" of the AI.
Lol it will eventually get cheaper. No way it stays at $3K a month. It will eventually be at the level of cents for the AI.
The limit is always energy efficiency.
relax dude whoever is saying jobs are being outsourced to India.. Its a lie, like US in India also there are no coding jobs. Indian managers never hire programmers who are over 4-5 years, people with 3 years experience are being recruited for Team Lead roles. Do you imagine?. Lot of changed over last 5 years. Managers will say why would we hire you when junior also can write the same code that you have written. I have faced this question many times during interview.
It is strange to compare a tractor to a farmer. The biggest difference from humans in ai is that automation is possible.
I love how people keep jumping on this AI replacing programmers thing. I use AI every day and I can tell you AI cannot code worth ship. Sure it can do basic coding but it cannot create a simple POJO that performs a complex function. It also cannot, in any shape or form, create code from pseudocode. I have tried a ton of different models and nothing. In fact the only way AI has worked for me is to assist in creating very basic code, and if supervised closely it can create some test code but I have little faith in that as well. I lost an entire day of coding because the previous programmer used AI to do 80% code completion and the code was so borked that the AI created test code only worked with the exact SAME input every single time. If you altered the input to the test code at all for the different values that can be absorbed by the code it failed. So when someone says AI can code better than a programmer... well that is just not true, it cannot take in complex prompts to create an application, no matter how detailed the prompt. It also cannot create code to log into certain access points, not wifi here but servers and DB's, it simply doesn't understand it. I even gave a local AI model my code that works 100% and it could not create a simple login to a DB to create a DB query that worked. I had to go in and tell it exactly what was wrong and after it fixed the code it broke other code. So no, again, AI cannot, at all, in any way fashion or form create complex, code and therefor it cannot replace programmers.
I wouldn't sound so absolute. AI is currently in its infancy. ChatGPT came out in 2022 and couldn't code at all. It's now 2024/2025... and there are claims that the unreleased model can code better than 99% of coders. Though that yet to be seen... if the claims are true, what level do you think it will be at a few years from now? Of course, no one wants AI to put humans out of work, but don't be so naive....
@@JayC216 the statement is oversimplified. Codeforce is not all programmers doing what they are paid to do. Writing code is just one part of that.
Not to mention the model was trained on the dataset.
@@drwhitewash It learned Codeforce today... and it'll learn something new tomorrow 🤷♂... and with the introduction of quantum computing and LLM's abilities to now think, reason, and create original concepts, not just pull information from a database... I think we're gonna be in for a big surprise in the next year or so. People are looking at AI's current limitations and dismissing it as if the advancement of AI hasn't been extremely rapid... and starting to pick up speed. That's like looking at a three-year old and saying "Ha! He'll never be better than me at anything!" as if that child won't learn and grow.
@@JayC216 to learn something new tomorrow, it will have to be trained on that. Public training set from these benchmarks was specifically included in training data.
Training of these huge models is expensive and will continue to be for the foreseeable future.
I don't know if it's starting to pick up speed or it has already started and had a few hiccups on the way already. Openai has its struggles and delays, that means the low hanging fruit is starting to be rare.
Don't forget the sigmoid curve.
However, today is not tomorrow. If things aren't on a trajectory but suddenly freeze and remain the same, then you have a valid point. However, your point is invalid.
for %80 of programmers, it is the end. Unless you are super smart. Average and below it is the end.
Next stop, Customers, good luck with that one AI.
Imagine how much work these AIs will give humanity to do!
When death happens, denial is the first emotion. The comment section is filled with it.
AI won't replace programmers till they figure out creating AI agents
This means that one developer is worth two. Thanks AI.
Who cares what today's costs are, costs come down very rapidly over time as newer versions come out
Haha, when I saw thumbnail, I thought "who is this muscular guy in a sexy white shirt? Let's click the video" and you got me. Good job getting my monkey-brain to pay attention haha :)
On-topic: hearingyour analogy about farming technology reducing the need for human farmers but this being good for the world, it makes me think about how scary change and unpredictability is. I can't help but think that if we, as a society, had more of a safety net such that people could adapt to change in our job landscape more easily, then there would be less anxiety.
eh I think it'll just make coders more productive, thus increasing their marginal revenue product, and making them more valuable. The ability to google stack overflow, and really good IDEs was a slightly bigger change but this is still a big deal.
You are cooked bro
If Open AI new model beats 99% of programmers then Open AI also beats itself and should go out of business with this release.
that exponential scale on the bottom. yeah i doubt a single prompt is worth that much for a single dev. or able to replace one cost effectively...
Programming as a career becomes more select. You can stay in denial but keep in mind this is the worse it will ever be! Microsoft CEO was right when he said that AI will replace the application layer to interact with data.
@@bigpickle8844 conversational interaction was never a good way to process a lot of information at once.
Can you explain last sentence in details, please?
OSI model
*database management software sadly leaving the room*
@@chillgamervids dbms is still needed, even more so :)
Additionally to everything else, you also need to store vector embeddings for the new kind of LLM based applications.
"What happens now?" you ask! Why do you ask a question to which you do not want to know the answer?
It's the end. No ostrich head syndrome please. Like calculator ended manual calculation, and many other jobs disappeared, it's for software engineering now
The boilerplate code that AI generates seems to impress non developers. It’s still miserably bad at slightly higher level concepts.