and the HR, and the directors, and the CEO and the VCs, and the consumers. the entire company is AI making bullshit products for another AI to consume, thus selling a lot of GPU for nVidia and actually producing 0 value for humans and using all of our power
The mass hiring on junior devs has been great for me, there is an endless amount of work that needs senior devs. Everyone hiring devs starts with low cost junior devs, which means they inevitably need senior devs.
Can't wait for the refactoring gigs to start pouring in. The "hey, we hired an AI to write our entire app and now it's a tutti frutti spaghetti codice, mamma mia, help us per favore" kinda gig.
@@ada0015See, how will AI entertain us with great code comments so that we can handle the pain of an undocumented obscure legacy codebase? Only humans write those. Maybe they should train an AI to make witty and angry comments. It really boost morals to see the salty and sarcastic comments, I would never have been able to push through without those about the windows API when I had to work with it and a legacy codebase.
Yes. This. To any manager reading this: It is a an absurd idea, forget it, it will not happen. When we have free food and free housing provided by AI, then we can talk about AI replacing developers. Stop getting trolled by marketing.
I have been a 'C' and 80x86 Assembler programmer for over forty years. I am 71 years old and my programming career started on al old ATARI 400 with a membrane keyboard and no external storage device. I do not have a degree; I am a self-taught coder. I have worked for government contractors supporting some of the biggest military contracts in the world. I have worked writing software for 911 dispatch centers. I have written data recovery software that has recovered thousands of client's data. I supported my family for forty years as a coder. With all that being said I have one observation to the question "Am I worried AI will replace coders?" BWAHAHAHAHAHAHAHAHAHA!!! Thank you, I feel much better now.
Humans are really good at sarcasm and understanding meanings where it's not even there 😂 like this what u just said. AI is probably never reaching that or will take super long time.
You guys are lucky! You at leest have managers that have ideas what need to be built. Managers here in NL are generally like: “we need to conform to these regulations.” Not even understanding those stupid leftwing laws themselves 😂 Or my recent contract: “you need to move to azure cloud.” And nobody knows where to start because there are also protocols as to how to do that in a bank, that nobody seems to know where they are. So is externals are then hired because we can hide behind the shield of ignorance and say: “sorry we didn’t know that these compliance rules were in place. Can you show us where they are?” And continue building whilst they are seeking their backlog of compliance processes. And by the time we have infra setup and migrated the systems. We spend another year getting the compliance in place which had already changed many times since we asked to get all the compliance rules for us😂 Because if you follow the corporate crap we all be wanking each other in a circle jerk and nothing gets done. And then they want to hire you on a perm contract and offer you shit. And you walk away.
80% of my last job was explaining to senior management why I shouldn't implement their idea, 20% was explaining to my teams why the resulting stupid idea was actually a better compromise.
Word. Or customers, who think they knows better, even after being told, it's a bad practise. I've experienced so many times, when a customers diden't listen to my feedback, i created what they demanded, they saw the product, and suddenly they agree, that it's a shitty idea.
As a person who failed to persuade a PM and than BAs of stupidity of their ideas and wasted 4 dev*years on in-house competitor to a unicorn level startup...
@@larsnobel4279At my job we work in a bit of an unorthodox way. But basically, when we go through what the client wants to do, we also have the requirement of knowing why they want the job done. Then the job turns into "does this fit their needs?" and you know the rest. But just knowing the why helps immensely. Because they have a perceived problem, but we might be able to solve it in a better way since we understand the system on a much deeper level.
As an experienced developer, I'm looking forward to the company dropping me for the AI - I'll get to set my rate as a contractor when they bring me back to fix the mess they get themselves into It's very telling that the big ideas are to replace the workers rather than picking the low hanging fruit and replacing the managers
Your right about replacing managers. Really the killer and most dystopian use case for AI will be something that monitors everything you do at work to evaluate you and rank you.
Tbh an AI mess is probably a full rewrite lol i'm not fixing that crap, thing i'm terrified about with this is all the juniors starting to use it and pushing commits under their name but generated with this crap, and that they cannot understand it later on, definitely puts new pressures on code review processes imo
I like how he skipped over the fact that he was constantly correcting, redirecting, and providing the URLs directly to Devin throughout the whole demo.
Yeah, like the "I broke ChatGPT" scripts... no.. you literally asked it to roleplay with you, loaded it up with the exact types of content and tone preambles you want, and it just branched off once or twice from there.. whoop-di-fkn-do. You didn't make "AI evil". There's nothing intelligent about it :P
Yeah and it actually took 1h for the zoo guy to get his answer. Given that he had the code that produced the simple bug before going to Devin, I feel like any developer with even few years of experience could have figured the issue out faster than in an hour. "Thanks Devin for saving a lot of time for me :D"
He only said some things that are true in the present, he has no fcking clue how AI systems will improve in the years to come. Don't be naive, stay prepared.
The Devin Loop: Devin, I need you to creat a build test pipeline. The pipeline should do the following: - Build Devin from source - Deploy Devin - Supply Devin with this text as a prompt - Log results
The problem is not the A.I. ... in fact artificial intelligence does not exist. Only the abuse of language exists! Hehehehe, which also kind of describes Large Language Models. They are copyright abuse. Problematic are gullible journalists and companies like Open(open? ... hahaha)AI and the absolute stoopid audience who eats every piece of BS they serve them. At the end, it's just another scam! History is that computer scientists were not over-exaggerating in the first place, but if there is a hype ... there is also money(funding!). You know the rest. Thank goodness for critical journalists like ThePrimeTime. But I really cannot stand that A.I.-BS anymore. At the first sight of the topic or other stoopid buzz-words worthy of an Elisabeth Holmes, I immediately block the channel/media/IP and mark it as ignored. All that is automated and done by an A.I. (forgive me, that joke had to be told!:P)
In the end, for anything more than a very basic tool, you're gonna have to write down exactly what you want, in precise language that mentions every single edge case you need. At that point, might just as well write it in code, because that's basically exactly that.
Yeah, if you already know what you want, you can just throw all the code overseas with exact requirements you want. There are tons of people out there who are ready to turn requirements to code nearly for free. The error rate will lower btw
No you don't have to write precisely what you want. Its not a regular software. Its an NLP AI. That's the point. You can iteratively chat with it. Watch their demo. There is a chat window. If your intention or communication wasn't correct and that produced unintended results, tell it.
I mean, HR wants us to have 40 years Java experience, to be a RISC programmer and to have spent 30 years before that building breadboards before they will consider you for the position. I will never forget being demanded why I only had 11 years experience in Java when the language was barely even 15 years old.
- 2800 years of experience with LLM fine-tuning of Egyptian hieroglyph holy scriptures - 56 years of experience with COBOL is a must. - Parsing HTML with Regular Expression Virtual Machines is a must. - Fluent Latin (800 B.C. dialect) is a must. - Willing to travel back in time (without a round-trip) is a must. - Willing to being exposed to 5000 mSv of ionizing radiation dosage during the time-traveling process. - Microsoft Office is a great plus.
Ok I just started coding like 3 weeks ago and I saw this Devin thing, got all depressed today, and I really needed this video. I’m having fun coding pong. I want to continue to learn and I want to make this a career. I hope I’m not too late to the party and this kind of helped me say “no I’m not too late.”
There are many here who are huffing the copium and will say jobs are just fine. Don't listen to them. Understand that the market is REALLY REALLY bad for inexperienced developers. If you like coding, continue to do so, but don't keep an expectation of a job. And if you do find a job, also try your luck in lottery because at that point, Lady Luck has kissed you on your forehead.
Im a programmer and I think you should work a trade. I’m working on joining to military rn lol. The industry may survive 10 years but it’s not worth gambling
Please don't give up. Programming is fun, and these tools can only replace the boring and repetitive part of our task. The cool part that is thinking, engeenering, and solving problems with software will keep with us.
Great take, you were spot on about it the video being for VC's. Also, Isin't it like the more the number of inputs, the more the abstraction, the more chances of errors ? Like they are googling, using their own terminal, I see maintaining Devin a job in itself . Honestly, the more the AI push comes in, the more it pushes me to get my basics stronger, gives you an added super power of understanding atleast a level of abstractions these things are built on.
Aa an old programmer from way back - now retired: to steal from Martin Niemoller: "First 4GLs came for my job and I did nothing Then the Japanese came for my job and I did nothing Then the outsourcing to India came for my job and I did nothing. And after 35 years I retired happily as a programmer." As Primeogen points out - far too many times - - the hard part of programming is understanding your application domain and being able to cast that into code. AI at this point can write code - but understand an application domain??? It will have impact on the job market - and coding skills will change. But historically, whenever we have found something to make programmers more efficient, we have just upped the bar on what we think a good program should be able to do. We will go through a few silly years where clueless management will try to kill off programmers. And they may succeed. But not with something like Devin
The problem is that even if Devin doesn’t write insightful or eye-opening code, that’s still enough to put people out of jobs I mean, chatgpt 4 already writes code like someone with 3 years experience and a brain injury, what will come next?
@@Dipj01 I see people making this analogy everywhere, and I feel dumber every time I read it. This kind of comment shows an obscene ignorance of survivorship bias. You lot will reference every time in recent history that a position has been automated without any comprehension that for every successful automation movement, there's 1,000 COBALs born into the world. It's on you to prove that AI can actually replace a professional developer, not on us to prove that it can't.
Waiting for that moment when AI generated code creates a catastrophe (I wish no fatalities), causing billions of money for a mega corp. And then the hype will settle down leading ultimately to burst the AI bubble.
No guys. Managers typically rely on devs which can filter out some shitty ideas and provide fast feedback. If Devid will jus do whatever they will told him to do, it will be total mess in a long run
As a guy working with communication and marketing I bet software engineers feel just about how I felt when I saw Chat-GPT for the first time about 2 years ago. Yeah it's not going to replace all the jobs - not even close. But everything happens at the margins, and if 5% of all jobs are lost to AI, wages go way down and finding a new job is super hard.
As a software developer (with my own company as a independent contractor, billing hourly + retainers), I am getting happier and happier about these developments. Like, kicking my feet in the air happy. Especially about other developers getting frustrated and throwing the towel in the ring or moving on to other careers (absolute music to my ears!). Not to spite them, no not at all! Its because I have this sneaking feeling, that for a short period (which could be years), we will have it pretty hard, but then for those who hang in there, its going to literally start rain money. The gross overestimation of what AI can do, is likely never going to crystalize at all (that's a bet that I am willing to place). And then all companies are going to rush back to hire actual developers to refactor all this BS code and write real stuff that lasts. And then I have been there all along; weathered the storm, networked, etc.
Do you mind giving me some advice? I have 4 years of experience, got laid off and it's been impossible to find a new dev role. I want to get into freelance work and being an independent contractor. Where should I start? My tech is JS/Node & cloud with both AWS and GCP. Just give me a general roadmap for what I should do, i'll learn any technology and do whatever is required to find clients. What should my first 6 months look like?
I agree with this in general. Though for companies to choose to replace developers, AI doesn't need to do 10x better than living developers, like a new product that replaces another product would typically require in order to succeed. The company adopting it just needs to survive in the market without those people, and make the product cost 10x less than the people it replaces. A crappier product that costs 10x less to produce is still enough to make a lot of execs pull the trigger.
The company would find that to be a big mistake, especially the longer time goes on. If people think we have problems with software now I'm excited to see the hell scape that software would be after years of shitty AI products.
@@hereandnow3156”Years” of shitty AI? If you’re saying the current state of AI will last for years or will be shitty for years, you’re pretty much saying you know more about AI and the future of AI than S.Altman, G.Hinton, D.Hassabis, I.Sutskever, A.Karpathy, M.Suleyman, D.Amodei, J.Huang, M.Gawdet, E.Mostaque, R.Kurzweil, and dozens of others at the forefront of this field. I don’t even think discussions and debates like this about Devin are relevant. It’s not whether Devin will replace software engineers. It’s whether OpenAI or MS build something similar with GPT-5 or GPT-6, or what Devin will be like in a year or two running Gemini Ultra 3.0 under the hood.
@@wonmoreminuteah yes, GPT-6, which will take 15 million nvidia GPUs and 2 years to train. (GPT-3 took 1k gpus and 30 days... GPT-4 took 25k gpus and 90 days... growth does not look promising)
@@deadbeef576 correct. and its assets get sold off to become someone else's problem or detritus. if a company can live long enough to let them extract a huge amount of value, for a large enough percentage of them (at least large enough for many employees of such companies to be wary, even if it's a single percent), the people making these decisions at the top will not care one bit.
Thank you PrimeTime. I had been kinda freaking out about this one in the afternoon but I said 'well lets see what Primeagen says' and you make so much sense. Thank you. You are the goat.
@@tomkac1991 my bad, I'll do some refactoring in all seriousness , I see a lot of denial and the truth will hit you a lot harder if you're not prepared for it ;)
One big problem if this AI thing actually ever takes off is that, over generations, it will cause people to lose the skill to analyze and solve problems, instead people will just rely on the Ai to solve problems for them. And when a time comes when it breaks down we'll have a huge catastrophe coming. Imagine even normal people no longer knowing how to repair a car, or how to grow food.
@@lucasjames8281Maybe, „driving cars“ would have been a better example due to the decade long hype of self-driving. „I am confident that we will have full self driving next year“.
You could pretend the same thing about rubbing sticks together to make fire. Just because someone's lighter ruins out of fluid doesn't mean the world will fall apart just because we didn't teach dudes how to deal with that shit. There will be an overabundance of safety nets.
the fact that they give the AI a terminal, code editor, and window is just so obviously a move to make it seem more "human" to me, signaling that this is some marketing thing for idiots who don't know anything about software and not for devs. Why the fuck does Devin need a little browser window to open the api docs? why not curl that shit directly into his prompt so he can use the information? why does Devin need an editor when he's a computer program that directly reads and writes to stdin/out? why isn't he just directly editing the code on disk?
One reason would be so the dev can debug and see how it arrived at its conclusion. You can actually scroll through the whole timeline of it's workspace while it works. Pretty cool.
You realize Devin doesn't have eyes, right? The web page is downloaded, and since it's already a web page, and you're using Devin in a browser... Showing it would be trivial. The browser, terminal, etc. mainly serve as visual aid for users and VC so it seems cooler
I’m a novice developer and whenever I’ve tried to use ai for something that had no documentation online the ai would be so full of mistakes it was unusable
You need to change the scope of your question, and reiterate. You will learn to make good use of the LLM. It actually takes practice and getting used to. It will work with even unknown concepts as it is capable of reasoning and following the pattern you provide. Some of the big mistakes people make using LLM: People give up too fast on the LLM and avoid iterating, and people also ask the wrong questions within the wrong scope, and do not give enough examples. Eventually you will learn and know exactly which questions the LLM will most likely be able to answer, and you can get to those very questions using the LLM as well. It's easy tool to underestimate because we are used to work in a certain worflow from years of experience without LLMs.
I've been looking for a job for almost 2 years and I have been ghosted towards the final interview stages for positions I was sure to get probably because of my time zone and location. After seeing Devin on Twitter, I closed my PC. I even took up a CS degree last year to get an upper hand and take time to get an understanding of concepts that other senior devs kept mentioning like memory management and so on(I could have learnt these solo but I still needed the CS degree to back me up). My productivity over time has been decreasing logarithmically. I feel cooked at this point and even if I was considering a career change, I put so much time into what I actually loved to do. Devin is awesome and helps speed up development but on the other hand, it reduces the chances of someone like me with so many disadvantages to make a living out of the career which stings. Just wanted to put that out, I'll get back to my grind and hopefully, things work out. I'm currently working on a project to solve a problem in my country with other devs in the same state that I believe would get me an upper hand, hopefully.
Listening to this while coding, concerned that I might be coding for the last times in history cause it's going away, and your soothing tone is just what I needed to be able to keep going (even if you end up being wrong). Thanks
1:10 This is a really fair take. Im currently going through Uni, studying CS, so all I am looking for is to learn, and not necessarily catch up with whatever level AI is currently at. I do not have 5+ years of experience, and yet AI seems to make me feel like thats what I need just to get started, else Im good for nothing. You shouldn’t need to be an expert to simply be able to contribute.
In my opinion, unless you need structured learning, college is a waste of energy and often times a major financial fuckup unless you have the government paying. As for you getting work as a SWE, focus on what you would enjoy building way more than the money. Don't be the person who becomes a lawyer to have a rich lifestyle. As for your experience, at the end of the day, business people value results. If you can deliver and solve a problem and give value, and they are not complete cheap asses, broke, or exploiting you, they will pay. Im a 15+ year dev who started as a pre-teen, with many hats, and I have 0 concerns about this and use AI near daily. Though im also doing very unique stuff and so another business lesson is specialize. The easier something is to do, unless there is another market factor, the more the price will be pressured down and commoditized. A good niche where you can deliver value will earn you a living.
@@scooterman30 Im in the same boat im about to graduate with masters in comp sci next year. Got decent scholarship, parents contributed some part, rest paid from part time jobs and was kinda scared with all these new AI coders popping up every other month. I even started learning aws cloud because of this and cleared foundation level cert exam. I was building a full-stack project and saw this AI coding stuff, got demotivated and thought is there even any point to build my own project when you can do it with AI but your comment gives me hope. Thanks
Focus on understanding the theory, the fundamentals. That’s the whole point of getting the degree anyway, and luckily, that’s not something that can be automated by software anyway!
@@spralwers theory is just another way of saying academia imho, and even that can be learned free if you really want to. I care more about practical and getting problems solved.
From my experience; in recent years it is more of an issue of adapting to ecosystem of companies I am working at than doing actual coding. I have no issue to code even very complex code that is required; but it is frustrating to get all the required info from management and/or mark all the "checkboxes" of workflow.
Right, and if it'll get exponentially better at solving ever-increasingly difficult programming problems, occupational competence will be decreasingly dependant about pure coding ability - the thing I feel mostly differentiates juniors from seniors? How is it that it's the well paid senior devs at highly paying positions that feel smug and irreplaceable?
If you fear about being replaced: The “I” on LLM stands for intelligence . If you train a LLM with data that says 1+1 is 3, it will always says that 1+1 is 3. It will never have a critical thinking and actual learn and think. Its just a word calculator. That’s all it is.
@@valerioharvey7289 Do you know the dictionaries where your phone learns the words that you use frequently and next time you write a similar sentence it suggests the words you already used? It's the same with LLMs but instead of short messages, they put it through thousands of internet articles etc.
@@valerioharvey7289 AI does not have the ability to discern lies or come up with ideas on its own, it simply gathers information and presents it to the user. The "AI" isn't smart enough to reject human bias.
@@user-og6hl6lv7p Buuut then you realize AI has the ability to discern that 1 is a single unit, and that 3 is three of those single units, therefore that 1+1 couldn't possibly ever be 3.
I feel like passing technical interviews is not surprising at all, it’s right up the alley of an AI, single problems that have been well documented and have plenty of examples on the internet, from what I have heard, most AIs (including Devin) eat shit when they have to code at scale for a large code base, some even claiming it has bugged out and deleted working code after trying to solve a bug in it. Not even to mention how, since AIs like to use boiler plate code off the internet, they tend to introduce a lot of security vulnerabilities. There are plenty of other things too that I’m too tired to think of.
Yeah exactly. I'm still waiting to see a workflow of one of these AI devs that's on an actual large real-world code base and not writing some simple code snippets in a vacuum. Then see it have to figure out where to put its new code in that existing code base, and what code that already exists to remove. It's a similar problem to self-driving cars, a person always has to be watching to make sure the thing doesn't drive into oncoming traffic.
Ah yes, can't wait for everyone to replace Juniors with AI and then figure out in 10+ years that there's no more Juniors to fill in the blanks of Seniors and AI Garbage (If AI doesn't receive info from humans it tends to bloat itself, and if it doesn't receive NEW content it won't grow and won't be able to deal with newer frameworks) Where did all the Juniors go? I thought we had a line of them lined up? Yeah, that was 5 years ago when people figured out they wouldn't be getting a job anytime soon because you went replacing everyone with AI and now everyone went into trades or manual labor. Edit: And yes, I've watched the video and agree on the fact that AI isn't going to replace jobs, but its not the experts or actual workers these AI companies have to convince, it's the managers and sales people who have no idea what the hell is going on and just want a bigger profit for themselves.
a) if ai keeps improving solely off the back of more data and not better algos - it's already doomed b) I understand the sentiment, but really? Trades and manual labour? Are there no other worthwhile majors?
As a former electrician it's kind of humorous every time I see software devs talk about switching to trades (seriously or ironically). Most people won't last a year as an apprentice, much less makes a career out of it just to get a salary of a janitor in return for busted knees and back by 40. Not the main point but compelled to comment
@@markerickson6320 Hey man I agree with point a) but it just seems that these companies just release a new version of an LLM and market it in some different way, I'm pretty sure there's progress behind closed doors, but without concrete evidence it just seems that it's just an LLM gold rush. About point b) and trades and manual labour, trades only take 2 years to learn for, considering a lot of people fear that AI will take their job, then time is on the short side and you want to go somewhere where AI can't automate, sure Surgeons, Doctors, Electronic Engineers, and others are better long term, but if people are going to get fired in mass even if they went to stuff like Computer Science becaues AI does it better, then you can't really go back to University for another 4 years or more if AI took your job, by the time you finish those degrees AI would've affected them as well. Community College is also a good option because it's faster, and time is the limit because if layoffs do happen, not everyone is going to have the luxury to move back with their parents. I understand that it's not the best, but its what I would personally do if I were caught in that situation.
@@aarholodian And as an electrician you've got one of the less back breaking jobs in the trades. I did months of work experience with my dad who was a carpenter and joiner, that shit is hard. Mega respect for tradies. I've watched carpetners singlehandled monkey climb up half built house frames and hold trusses in place with one hand and using a nailgun with the other, genuinely most office working people could never last a week in that kind of job.
You could argue one developer could discuss the requirements and Devin would do the coding. With the speed an AI codes this could replace thousands of developers. That said I don't think AI will ever be relevant as a developer unless we create an AI that thinks as a human and understands context and if that does happen then we don't have to care about jobs anyways since AI could supply us with all we need.
Most useful thing in the first demo is that initial plan it creates and collecting the documentation. I approaches something i would pay for because that can really save some engineering time.
i hate how LLMs are put in the work we like doing(and as devs the code is the thing we are good at not the rest), just put LLMs in the tech stuff we dont like doing. Make it write documentation for an existing legacy system, make it prepare a cool presentation for the executives for the last sprint completed, make it flag commit messages when they are not descriptive enough(or somehow enhance bad commit messages). Make it look-up obscure bugs that might be found in code submitted for a review.
Even if Devin could do nothing else but read the documentation, read the code and submit pull requests for fixes to documentation, it would be great. Assuming it's cheap enough to use. The demo to read a blog post and repeat the same thing with ControlNet for about $750 is not a threat to any human developer.
Thats true but you need to consider the fact that problems are not uniformly distributed. Most of the SWE problems that businesses have are easy. Some businesses use only spreadsheets, forms and other already existing tools. Some need small customizations. Very hard problems occur very rarely. It still means that Devin might affect thousands for SWE job positions.
@@darekmistrz4364 yeah that's very much true but I do think low level developer are worthy of replacing, im fact I think they actually should be in my opinion because of the learn software earn money society around software engineering this field is full with developer that earns a lot without actually putting any value to the company, which I think they should be replaced.
I spend time every week testing AI tools to see if any can improve my workflow. The most productive tool for my case is LM Studio running Mixtral 8x7B Instruct. It's a brilliant example of how shit the benchmarks are or how much overfit there is towards the benchmark. I ask it to write a function that I've written a stub for while I keep going. When I'm done with my task I check if it's function runs or if it has to to re-generate. My 4090 goes hard and every 8 times it produces a junior level function. Edit: Worth mentioning that I work in the company that I co-founded and own more than a third of so my incentives is maximum productivity.
That video is such a marketing tool. All of this just reeks of a startup that really wants to get some kind of market share. It's crazy how much of this comes from the crypto playbook. I've been in the industry for 25 years, I'm not worried about Devin taking my job anytime soon.
There's new hype every few years and it rarely materializes. I was there for Enterprise Java Beans, SOAP and a few other "life-changing" silver bullets. Just like Prime said, this is just to get money from investors.
@@ArturdeSousaRocha yup, exactly. That said, even comparing to how people marketed things back in the java days, it was less insidious compared to how. With crypto, so many people just stopped caring about pretending to have any ethics.
@@btm1 True, every once in a while there comes a disruptive solution that upends the status quo. Usually, though, it's not accompanied by hype. Things like WWW or social media just slowly gained traction and only *then* businesses started piggybacking some hype on them.
@@btm1 not being naive, I'm being realistic. I've worked on AI myself and I've read many of the research papers, there is a lot of evidence to point to the fact that we are not going to hit a point where these models are going to exponentially get better. Probably the biggest thing is that they're going to run out of usable data by 2025 which means that for the models to get better, they will need to improve the algorithms which I'm sure will happen but not at the pace they have been recently. There was a research paper on co-pilot, where ever since it's inception, code quality in general has gone down. This makes sense because these models don't really select for quality code as much as they exist to just solve the problem by any means possible. Solving a problem and creating something that can be maintained for a long period of time are two different and distinct issues at the end of the day. For non trivial usecases, the code needs to be maintainable or else even these AI workers will succumb to technical debt. To me the most compelling piece of evidence are the numerous examples that show that these models don't really understand anything about the training data. Take for example the recent snafu with Gemini where it declared c++ unsafe for users under the age of 18. The fact that it couldn't figure out the context of the term unsafe with regards to programming versus unsafe in a general sense proves that it doesn't understand language as much as it just predicts patterns. Even prior to this Gemini incident, there was a paper on alphago where the researchers showed that they could consistently beat the algorithm using one of the most foundational concepts of the game. If the model truly understood the game of go, then it wouldn't have been susceptible to such tactics.
Did Primeagen predicted SORA or GPT-4? He may say some things that are true in the present, but he has no fucking clue how much will AI evolve in the next few years. Hell, even the experts were surprised by how much progress was done in the last years and the emerging properties that appeared in AI systems.
@@btm1 i'm sorry you're so insecure about your pov you need to crawl comments in search of people being hopeful about future of software of development to tell them how wrong they are
15:15 I'm extremely curious too to see how that turns out. I was able to be in a flow state for a long time working on a non-programming side project where I mainly used chatGPT and google sheets, but chatGPT wasn't there as a solver but as an unblocker of sorts. This was something creative and while AI can't give you anything particularly creative, seeing something *wrong* makes it easier to figure out what you actually think is *right* when you feel stuck or out of inspiration.
@@IsaacSchubert I love both, but when I'm stuck on one side LLMs vhanges the one way path into a complete road. Usually you can only switch from editing to writing, LLMs unlock the other direction.
Implementing a solution is just a part of my job. I spend a lot of my time working through design, pushing back on managers, and communicating with stakeholders. Obviously coding is important, but being in SWE is so much more than that. If anything, I'm hoping AI will save on engineering time and allow me to focus on the human and design-facing elements of work. Thank you for this refreshing take, I'm so tired of all the AI grifters and fear-mongering online recently.
As an developer with 4 years of experience, I am extremely worried. The market is such a mess that I can barely get an interview and when I do, and I do well on the interview, they tell me that I have been outperformed by someone with WAY more experience than me.
it took me a while to realize how over-hyped the generative-text goldrush is. the main problem I see is that it completely de-emphasized reinforcement learning and other forms of training/pattern recognition. as a result, AI research and funding is actually slowed while a pointless joke technology receives 95% of investment
Finally, a super based take on all the AI fear mongering to attract venture capital. Really appreciate this video! It is the first and only video I have seen on RUclips which gives a realistic perspective on the AI gold digging.
We really are in the end-times Devin (as of today) is nothing more than a few tweets and some YT videos. There is no way to actually use it. It's 2024 - this is considered a groundbreaking achievement in software engineering. Some tweets, a promotional video, and a blog. That's it ! Let that sink in. I think I will wait for having at least half an hour of using it before forming an opinion.
@@sid4579 we have better infrasructure or machine than you doing soldiering, and a lot of simulation software to simulate semiconductor, i would say any engineering job is AI anyway human are just the front end you have no control over science anymore.
Hello to anyone reading this. I'm currently on the verge of finishing a high school with an emphasis on computer science (I'm from Europe). I've participated at coding Olympiads, finished CS50 and The Odin Project, and dabbled in game development. A couple of years ago, if anyone were to ask me what I'm planning to pursue in the future, I would have probably said some kind of software engineering job. However, considering the current state we are in, I'm not so sure I have the security of knowing that my skills will even be relevant anymore in 4 years. I've been thinking about trying to put less of an emphasis on coding, to explore other parts of computer science, such as cybersecurity or embedded systems. Maybe by the time I finish university those markets would not be as saturated as a frontend developer for example. I would love to hear the thoughts of anyone with more experience, as I feel quite lost in this field at the moment. Thank you for spending the time to read this!
The discouraging thing is that it's really hard to predict what AI will actually be able to accomplish in a few years. It could end up being a strong booster for many people or it might be wholly snatching up jobs we don't even consider now. If it's the latter though, we will have societal problems on a scale that they can't be solved by individuals.. and at that point we need big changes in the society or we will be heading to a revolt.
"AI will probably be smarter than any single human next year" has the same energy as "In ~2 years [~2018], summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"
Some people think that fast progress is equivalent to exponential. We could definitely end up in a Moore's law scenario where we have rapid progress that eventually stalls out due to technical limitations or whatever.
Super wishful thinking, public AGI in months to 3 years max, ASI one year after that, Sam Altman will get trillions for a system that gets us to ASI and AGI achieved internally already at 99% probability. Trying to slowly drip technology to not induce mass panic , but eventually will be forced to let some beast out (even it its tamed down and not AGI will be societal changing) . Hardware won't be a limiting factor for very long. Groq chips are better than Nvidias, Quantum Computers are around the corner, many new technologies are on the brink of being released , and Devin is dumb compared to what exists internally. Even Claude3 seems smarter than Devin. The technology we see is in the infancy stage, it'll explode within 2 years and be fully matured. Thinking linearly will hurt you more nowadays. We can't limit ASI , if it's really millions or billions of times smarter than us, it will just create new hardware or technology that will be exponentially better each time, and even if there's a limit on it , it would be further than humans can comprehend. They will be able to bend physics to their will and master all the laws of the universe. Discover things humans would never discover too. We just are looking at hardware and software, maybe there's flexiware/middleware/physioware/bioware/ or something inbetween or in other dimensions that we can't perceive. Start of AI history, human's final invention.
@@phen-themoogle7651quantum computers are not around the corner, this would require an experimental (and probably theoretical) solution to quantum decoherence
You can basically already see the effects of this with e.g. ChatGPT. Try to get it to write a working Minecraft Datapack. It can't, because the available documentation is not only horrible but also out of date.
been a software engineer for 8 years, i use AI tools all the time. you are correct, it isn't taking SWE jobs anytime soon. ill be worried when an AI can do my job, but right now none can do that. in fact i have to struggle with current AI capabilities to keep on accelerating my current workflow. so I agree with you that it has to 100x the amount of capability it has right now to even become close to replacing jobs
Agreed, I like AI, but I certainly see all its pitfalls. My stance is, as long as it's based on 0s and 1s, we aren't getting generalised AI/advanced AI.
I have to watch Prime on 2x speed, which is hard bcs he talks so fast but man... there is so much mumbling and bullshitting around important, genuine good takes
I think this is actually a good thing because if you want to stay in this industry you have to continuously working and learning and try to improve, unless some company makes an "magical ai" that is able to replace all the software engineers then I might problem but the nature of "if you don't work hard you will be replaced" isn't actually that bad. actually its good to see front end dev getting replaced.
Did some research looking into the team at Cognition Labs and its basically all mathematics and iOi Olympiad gold medalists and codeforces grandmasters. LOL these people are literally at the top of the top of their game in Olympics mathematics and competitive programming. I do believe they will achieve their goals with Devin. Give them 1 or 2 more years and it may shock people, specially the ones in denial right now.
@@darkbird2724 soooo close, but still not there. It seems that the final self driving is 70% finished but the last 30% seem to be extremely difficult. Maybe Devin can ... or not?
I am so surprised that there are people out there who are still surprised by this as if no one knew that it's already available, and that it will be updated in two months that it will just be impossible to make flaws. It's such old news....
This guy must not have room in his fridge for food when it's filled with so many bottles of the COPIUM he's gotta be drinking too not see where this is leading 😂😂😂
@@BeerMoneyforTokyo yeah but I think what triggers me the most, and I think that for Prime too, is these companies putting out these bold statements to get VC money without thinking one moment about the consequences of their claims on the rest of society. Specially when the claims are highly dubious, like these
That point about flow state is very very true. I’m at that 6 years of experience point now so I’m not scared too much by these AIs and often I find myself using gpt more as a colleague whom I talk about an issue with, rather than actually copy pasting actual code.
Same. I use ChatGPT to help me code everyday, but I don't think I've ever once copy-pasted any code directly from the prompt outputs. I use it more as a conceptual research assistant and glorified search engine than anything else. If I do have it write code, it's so that I can better understand the language/framework I'm working with, and then I'll actually write it myself for the project I'm working on. tl;dr it's a programming assistant, not a coding assistant. Works best if you already know how to code. Important distinction right there.
The value is the way it plans steps and applies changes to existing applications. The actual value is not only in the code it generates, but the automation of steps which can be applied in more use cases. Perhaps in the future you can prompt it something, then it can generate the automated steps to achieve a task and do the task. For example, you can ask it to integrate with a smart home service it has never integrated with, it can automate the integration by creating the code and then perform the task. That would extend it's capabilities beyond code generation.
I'm new to programming just like a lot of people and I'm glad him and other youtubers are addressing this and making sure newbies know its ok to keep learning and not get discouraged. F**k you Devin
5 years ago it was at 0-1%, 5 years from now it will be higher. Not being a doomsayer, just sharing my observations that technology _always_ progresses until we all die.
The other big problems come with questions like "What happens when it fails?" Does it just put in a non-functional best guess garbage code into your codebase? A big issue with ChatGPT is it rarely says "I can't do that", it just makes up some crap that looks like a solution but fails. Imagine a codebase polluted with nonfunctional or buggy code like that. Great it may do a good job 14% of the time, but who is cleaning up its failures?
I think the prime is missing this point; the way he looking at it is like, the day we reach full llm automation, is when SE start losing jobs. I think he is ignoring the fact that LLM/AI just increases the competition in the market, and that how SE will lose their roles.
Right. I see a lot of copium here, but Big Tech has already made their stance clear on this issue with layoffs. Even if AI can't replace programmers yet, companies are getting leaner to invest in AI and are actively looking to layoff where they can.
It's not a percentage thing. AI can't and likely wont be able to do what even junior devs can do, because it's not a matter of being able to regurgitate variations of pre-existing solutions, it's being able to logically understand WHY you're writing the code you're writing. All that will happen is in 5 years, companies that rely fully on AI will be screwed and the companies that hired and developed people will succeed. The entire industry isn't going to collectively decide to stop hiring all at once.
@@DeveStarr You dont understand, nobody buys AI generated stuff. Ever seen how people flip their shit about asset swaps in games made with Unity? People dont want AI synthesized art or voices in their games, the mindset is "why do have to pay 70 dollars for something that cost nothing to produce?"
Im a first year college software engineer student, and this whole devin thing is making me very anxious lol. I dont want to quit and stop learning, so what kind of programming thing should I try to study now? how can I find the most important area to go into? I was thinking of stuyding how to work along with these new AI's but how can I look for that to study lol?
Devin is actually controlling the company, hiring cringe people to do "believable P. R.", and software engineers to work on what he's supposed to work on, and in the meanwhile he's working on his own code.
If it's easy to replace developers, imagine how easy it might be to replace managers
and the HR, and the directors, and the CEO and the VCs, and the consumers. the entire company is AI making bullshit products for another AI to consume, thus selling a lot of GPU for nVidia and actually producing 0 value for humans and using all of our power
maybe we should just cut to the endgame and use AI to replace Wall Street
@@monad_tcp over for brokercels
@@monad_tcp it is already like that
Right now I don't even need a manager. I just need a task board with tasks and the contact details of the people for whom I write the software.
I work with a team of devs who have no idea what the hell they're doing. I don't fear being replaced. I fear being lonely.
ill join you. i agree 100%
The mass hiring on junior devs has been great for me, there is an endless amount of work that needs senior devs.
Everyone hiring devs starts with low cost junior devs, which means they inevitably need senior devs.
@@CDeCollibus awwwww ty!
Are you hiring??
@@adrianparker3116on the other hand, if most companies stop hiring junior devs, senior devs stop being created which would become a long term problem
Can't wait for the refactoring gigs to start pouring in. The "hey, we hired an AI to write our entire app and now it's a tutti frutti spaghetti codice, mamma mia, help us per favore" kinda gig.
All I will reply is: Pasta la vista
lol how do you come up with those sentence. pure gold
isnt it "por favor"?
@@blindshellvideos that's spanish. But his sentence is referencing italian..very similar languages though!
@@ada0015See, how will AI entertain us with great code comments so that we can handle the pain of an undocumented obscure legacy codebase?
Only humans write those.
Maybe they should train an AI to make witty and angry comments.
It really boost morals to see the salty and sarcastic comments, I would never have been able to push through without those about the windows API when I had to work with it and a legacy codebase.
The problem isn't that we are close to being replaced by AI. The problem is that we are close to hiring managers thinking we can be replaced by AI.
The owners of companies ultimately decide how many employees are needed.
God tier take
This.
Yes. This. To any manager reading this:
It is a an absurd idea, forget it, it will not happen. When we have free food and free housing provided by AI, then we can talk about AI replacing developers. Stop getting trolled by marketing.
What's done is done. Pandora's box has been open, and we can't stop progress.
I have been a 'C' and 80x86 Assembler programmer for over forty years. I am 71 years old and my programming career started on al old ATARI 400 with a membrane keyboard and no external storage device.
I do not have a degree; I am a self-taught coder.
I have worked for government contractors supporting some of the biggest military contracts in the world. I have worked writing software for 911 dispatch centers. I have written data recovery software that has recovered thousands of client's data.
I supported my family for forty years as a coder.
With all that being said I have one observation to the question "Am I worried AI will replace coders?"
BWAHAHAHAHAHAHAHAHAHA!!!
Thank you, I feel much better now.
damn....grandpa got a resume
Fireship fills me with existential dread, and you bring me back to earth. Thanks prime!
I came from the same video and I couldn´t put it with better words.
lmao just got here from that fireship video
You misspelled "herd mentality" 😂
Programmed to dread living life?
Fireship is being sarcastic lol
If the AI dev builds exactly what the manager asks for then most businesses would fall apart in a few months.
Lol so true
IMO this will repeat journey of SQL
Humans are really good at sarcasm and understanding meanings where it's not even there 😂 like this what u just said. AI is probably never reaching that or will take super long time.
But the AI can give Business impact of those actions
You guys are lucky! You at leest have managers that have ideas what need to be built. Managers here in NL are generally like: “we need to conform to these regulations.” Not even understanding those stupid leftwing laws themselves 😂
Or my recent contract: “you need to move to azure cloud.” And nobody knows where to start because there are also protocols as to how to do that in a bank, that nobody seems to know where they are. So is externals are then hired because we can hide behind the shield of ignorance and say: “sorry we didn’t know that these compliance rules were in place. Can you show us where they are?” And continue building whilst they are seeking their backlog of compliance processes. And by the time we have infra setup and migrated the systems. We spend another year getting the compliance in place which had already changed many times since we asked to get all the compliance rules for us😂
Because if you follow the corporate crap we all be wanking each other in a circle jerk and nothing gets done. And then they want to hire you on a perm contract and offer you shit. And you walk away.
80% of my last job was explaining to senior management why I shouldn't implement their idea, 20% was explaining to my teams why the resulting stupid idea was actually a better compromise.
Word. Or customers, who think they knows better, even after being told, it's a bad practise. I've experienced so many times, when a customers diden't listen to my feedback, i created what they demanded, they saw the product, and suddenly they agree, that it's a shitty idea.
@@larsnobel4279I lost this passion in 2 years… they want an ugly app - I make an ugly app, play mortal kombat and eat free pizza
I feel this comment in my bones.
As a person who failed to persuade a PM and than BAs of stupidity of their ideas and wasted 4 dev*years on in-house competitor to a unicorn level startup...
@@larsnobel4279At my job we work in a bit of an unorthodox way. But basically, when we go through what the client wants to do, we also have the requirement of knowing why they want the job done. Then the job turns into "does this fit their needs?" and you know the rest. But just knowing the why helps immensely. Because they have a perceived problem, but we might be able to solve it in a better way since we understand the system on a much deeper level.
I'm a simple man. I hear "Chat Jippity" and I laugh. Every time.
I'm legally changing my name to Devin, that way when I switched jobs the recruiter will think they are employing the actual Devin. Checkmate AI.
haha, good ideea
😂😂😂
unfortunately this wont work because Devin will be writing the recruitment software....
@@thinkingcitizenyeah, devin will be careful to check for its own ID number
works UNTIL they employ ai to hire ai . mate
As an experienced developer, I'm looking forward to the company dropping me for the AI - I'll get to set my rate as a contractor when they bring me back to fix the mess they get themselves into
It's very telling that the big ideas are to replace the workers rather than picking the low hanging fruit and replacing the managers
Yeah because we sure did see a surge of farmer workers returning to the fields after the farming machines that took over their jobs failed.
@@YeeLeeHaw Yeah, totally applicable and sound analogy
@@YeeLeeHaw huge dipshit joined the chat
Your right about replacing managers. Really the killer and most dystopian use case for AI will be something that monitors everything you do at work to evaluate you and rank you.
Tbh an AI mess is probably a full rewrite lol i'm not fixing that crap, thing i'm terrified about with this is all the juniors starting to use it and pushing commits under their name but generated with this crap, and that they cannot understand it later on, definitely puts new pressures on code review processes imo
I like how he skipped over the fact that he was constantly correcting, redirecting, and providing the URLs directly to Devin throughout the whole demo.
Yeah, like the "I broke ChatGPT" scripts... no.. you literally asked it to roleplay with you, loaded it up with the exact types of content and tone preambles you want, and it just branched off once or twice from there.. whoop-di-fkn-do. You didn't make "AI evil". There's nothing intelligent about it :P
Yeah and it actually took 1h for the zoo guy to get his answer. Given that he had the code that produced the simple bug before going to Devin, I feel like any developer with even few years of experience could have figured the issue out faster than in an hour. "Thanks Devin for saving a lot of time for me :D"
This video actually restores my emotional breakdown
And then the next day, this guy gets hired by Cognition, gets paid 6 digits, and then he teaches the Ai how to be a true professional
AI will eventually replace them. You remember how shitty websites were 10 years ago?
He only said some things that are true in the present, he has no fcking clue how AI systems will improve in the years to come. Don't be naive, stay prepared.
Agreed@@btm1
@@btm1yah you know they said these things before
The Devin Loop:
Devin, I need you to creat a build test pipeline. The pipeline should do the following:
- Build Devin from source
- Deploy Devin
- Supply Devin with this text as a prompt
- Log results
Devin, create a software like you but better.
Then give it the same prompt.
The problem is not the A.I. ... in fact artificial intelligence does not exist. Only the abuse of language exists! Hehehehe, which also kind of describes Large Language Models. They are copyright abuse.
Problematic are gullible journalists and companies like Open(open? ... hahaha)AI and the absolute stoopid audience who eats every piece of BS they serve them. At the end, it's just another scam! History is that computer scientists were not over-exaggerating in the first place, but if there is a hype ... there is also money(funding!). You know the rest. Thank goodness for critical journalists like ThePrimeTime. But I really cannot stand that A.I.-BS anymore. At the first sight of the topic or other stoopid buzz-words worthy of an Elisabeth Holmes, I immediately block the channel/media/IP and mark it as ignored. All that is automated and done by an A.I. (forgive me, that joke had to be told!:P)
@@midoevil7 actually yes, like "[loop Devin] with test Devin on X benchmark until it reaches 90% success"
@@Julzaa @midoevil7 @ryangrogan6839
f*ck. these are hilarious 🤣
Devin: Recursive prompt detected. Programming halted. GTFOutta here!
Devin is like an intern with 15 years of internship experience.
In the end, for anything more than a very basic tool, you're gonna have to write down exactly what you want, in precise language that mentions every single edge case you need.
At that point, might just as well write it in code, because that's basically exactly that.
Whoa.jpg
Yeah, if you already know what you want, you can just throw all the code overseas with exact requirements you want. There are tons of people out there who are ready to turn requirements to code nearly for free. The error rate will lower btw
it's basically prototyping.
"hey Devin, write me a code that does stuff" "see guys devin is useless is just bs hype"
No you don't have to write precisely what you want. Its not a regular software. Its an NLP AI. That's the point. You can iteratively chat with it. Watch their demo. There is a chat window. If your intention or communication wasn't correct and that produced unintended results, tell it.
"For every complex problem there is an answer that is clear, simple, and wrong."
Might get past a dev interview, but will fail to get past HR. It has no degrees or certificates and no work experience.
I mean, HR wants us to have 40 years Java experience, to be a RISC programmer and to have spent 30 years before that building breadboards before they will consider you for the position.
I will never forget being demanded why I only had 11 years experience in Java when the language was barely even 15 years old.
@@Rexhunterj HR is the ultimate boss for us all 😔
@@Rexhunterj 20 years in react required or don't bother applying. lol
@@Rexhunterj @litpath3633 what if it can simulate 40 years in a day :p
- 2800 years of experience with LLM fine-tuning of Egyptian hieroglyph holy scriptures
- 56 years of experience with COBOL is a must.
- Parsing HTML with Regular Expression Virtual Machines is a must.
- Fluent Latin (800 B.C. dialect) is a must.
- Willing to travel back in time (without a round-trip) is a must.
- Willing to being exposed to 5000 mSv of ionizing radiation dosage during the time-traveling process.
- Microsoft Office is a great plus.
Ok I just started coding like 3 weeks ago and I saw this Devin thing, got all depressed today, and I really needed this video. I’m having fun coding pong. I want to continue to learn and I want to make this a career. I hope I’m not too late to the party and this kind of helped me say “no I’m not too late.”
There are many here who are huffing the copium and will say jobs are just fine. Don't listen to them. Understand that the market is REALLY REALLY bad for inexperienced developers. If you like coding, continue to do so, but don't keep an expectation of a job. And if you do find a job, also try your luck in lottery because at that point, Lady Luck has kissed you on your forehead.
Im a programmer and I think you should work a trade. I’m working on joining to military rn lol. The industry may survive 10 years but it’s not worth gambling
Please don't give up. Programming is fun, and these tools can only replace the boring and repetitive part of our task. The cool part that is thinking, engeenering, and solving problems with software will keep with us.
your not to late to become a plumber
And no surprise, turns out Devin is just a big scam with it’s demo and it was just heavily overhyped
Great take, you were spot on about it the video being for VC's. Also, Isin't it like the more the number of inputs, the more the abstraction, the more chances of errors ? Like they are googling, using their own terminal, I see maintaining Devin a job in itself . Honestly, the more the AI push comes in, the more it pushes me to get my basics stronger, gives you an added super power of understanding atleast a level of abstractions these things are built on.
Aa an old programmer from way back - now retired: to steal from Martin Niemoller:
"First 4GLs came for my job and I did nothing
Then the Japanese came for my job and I did nothing
Then the outsourcing to India came for my job and I did nothing.
And after 35 years I retired happily as a programmer."
As Primeogen points out - far too many times - - the hard part of programming is understanding your application domain and
being able to cast that into code. AI at this point can write code - but understand an application domain???
It will have impact on the job market - and coding skills will change. But historically, whenever we have found something to make programmers more efficient, we have just upped the bar on what we think a good program should be able to do.
We will go through a few silly years where clueless management will try to kill off programmers. And they may succeed. But not with something like Devin
It definitely knows all the domains. Understanding is a different question. But we will get there
The problem is that even if Devin doesn’t write insightful or eye-opening code, that’s still enough to put people out of jobs
I mean, chatgpt 4 already writes code like someone with 3 years experience and a brain injury, what will come next?
That's like a farmer saying his job won't be automated because it hasn't yet
@@Dipj01 I see people making this analogy everywhere, and I feel dumber every time I read it. This kind of comment shows an obscene ignorance of survivorship bias. You lot will reference every time in recent history that a position has been automated without any comprehension that for every successful automation movement, there's 1,000 COBALs born into the world. It's on you to prove that AI can actually replace a professional developer, not on us to prove that it can't.
Waiting for that moment when AI generated code creates a catastrophe (I wish no fatalities), causing billions of money for a mega corp. And then the hype will settle down leading ultimately to burst the AI bubble.
No guys. Managers typically rely on devs which can filter out some shitty ideas and provide fast feedback. If Devid will jus do whatever they will told him to do, it will be total mess in a long run
Instead of a team of devs, it will just be that one dev reporting to the PM. Like the supervisor at the assembly line
@@TomNook. But without coding he will morph into manager )
@@32zim32 so we don't need manegers now. Great
Funny how devs not doing what you told them to do is a feature now.
@@user-df5ym9dv5galways has been
I appreciate the sanity and attempt to counter a lot of the FUD clickbait out there. I also appreciate the MS-Paint Total Recall bewbs.
Genuinely trying.
Genuinely awesome @@ThePrimeTimeagen
We are on the same vibe.
@@ThePrimeTimeagenI think your issue was adding the third one.
@@ThePrimeTimeagenalthough to be honest, you sound very annoying when you talk, and I think you have ADD.
As a guy working with communication and marketing I bet software engineers feel just about how I felt when I saw Chat-GPT for the first time about 2 years ago. Yeah it's not going to replace all the jobs - not even close. But everything happens at the margins, and if 5% of all jobs are lost to AI, wages go way down and finding a new job is super hard.
Gonna re-watch this video every time I find myself rethinking my life choices. Thank you prime
Maybe programmimg was the friends we made along the way
I threw up a little
@@zeuglcockatrice4633 LOL legendary reply 🤣🤣
So, 0?
very original joke buddy 👏
it's not like every video about Devin has this joke
You guys have friends? 😮
As a software developer (with my own company as a independent contractor, billing hourly + retainers), I am getting happier and happier about these developments. Like, kicking my feet in the air happy. Especially about other developers getting frustrated and throwing the towel in the ring or moving on to other careers (absolute music to my ears!). Not to spite them, no not at all! Its because I have this sneaking feeling, that for a short period (which could be years), we will have it pretty hard, but then for those who hang in there, its going to literally start rain money. The gross overestimation of what AI can do, is likely never going to crystalize at all (that's a bet that I am willing to place). And then all companies are going to rush back to hire actual developers to refactor all this BS code and write real stuff that lasts. And then I have been there all along; weathered the storm, networked, etc.
Its not even AI, its just a language model.
right
@@deadbeef576Language models are AI.
Copium
Do you mind giving me some advice? I have 4 years of experience, got laid off and it's been impossible to find a new dev role. I want to get into freelance work and being an independent contractor. Where should I start? My tech is JS/Node & cloud with both AWS and GCP.
Just give me a general roadmap for what I should do, i'll learn any technology and do whatever is required to find clients. What should my first 6 months look like?
I agree with this in general. Though for companies to choose to replace developers, AI doesn't need to do 10x better than living developers, like a new product that replaces another product would typically require in order to succeed. The company adopting it just needs to survive in the market without those people, and make the product cost 10x less than the people it replaces. A crappier product that costs 10x less to produce is still enough to make a lot of execs pull the trigger.
And then the company dies
The company would find that to be a big mistake, especially the longer time goes on. If people think we have problems with software now I'm excited to see the hell scape that software would be after years of shitty AI products.
@@hereandnow3156”Years” of shitty AI? If you’re saying the current state of AI will last for years or will be shitty for years, you’re pretty much saying you know more about AI and the future of AI than S.Altman, G.Hinton, D.Hassabis, I.Sutskever, A.Karpathy, M.Suleyman, D.Amodei, J.Huang, M.Gawdet, E.Mostaque, R.Kurzweil, and dozens of others at the forefront of this field.
I don’t even think discussions and debates like this about Devin are relevant. It’s not whether Devin will replace software engineers. It’s whether OpenAI or MS build something similar with GPT-5 or GPT-6, or what Devin will be like in a year or two running Gemini Ultra 3.0 under the hood.
@@wonmoreminuteah yes, GPT-6, which will take 15 million nvidia GPUs and 2 years to train. (GPT-3 took 1k gpus and 30 days... GPT-4 took 25k gpus and 90 days... growth does not look promising)
@@deadbeef576 correct. and its assets get sold off to become someone else's problem or detritus. if a company can live long enough to let them extract a huge amount of value, for a large enough percentage of them (at least large enough for many employees of such companies to be wary, even if it's a single percent), the people making these decisions at the top will not care one bit.
Thank you PrimeTime. I had been kinda freaking out about this one in the afternoon but I said 'well lets see what Primeagen says' and you make so much sense. Thank you. You are the goat.
I understand the cope but the truth is he can't predict how AI will evolve in the next years. Stay prepared.
@@btm1DRY. You're posting that everywhere. "Stay prepared" 😅
@@tomkac1991 my bad, I'll do some refactoring
in all seriousness , I see a lot of denial and the truth will hit you a lot harder if you're not prepared for it ;)
One big problem if this AI thing actually ever takes off is that, over generations, it will cause people to lose the skill to analyze and solve problems, instead people will just rely on the Ai to solve problems for them. And when a time comes when it breaks down we'll have a huge catastrophe coming. Imagine even normal people no longer knowing how to repair a car, or how to grow food.
Imagine it ? Like 99.9% of people don’t know how to do the last things you mentioned
@@lucasjames8281😂😂😂 we’re already there
@@lucasjames8281Maybe, „driving cars“ would have been a better example due to the decade long hype of self-driving. „I am confident that we will have full self driving next year“.
@@BeerMoneyforTokyo No correlation to self driving and also doesn‘t make the advances in self-driving care more impressive.
You could pretend the same thing about rubbing sticks together to make fire. Just because someone's lighter ruins out of fluid doesn't mean the world will fall apart just because we didn't teach dudes how to deal with that shit. There will be an overabundance of safety nets.
the fact that they give the AI a terminal, code editor, and window is just so obviously a move to make it seem more "human" to me, signaling that this is some marketing thing for idiots who don't know anything about software and not for devs. Why the fuck does Devin need a little browser window to open the api docs? why not curl that shit directly into his prompt so he can use the information? why does Devin need an editor when he's a computer program that directly reads and writes to stdin/out? why isn't he just directly editing the code on disk?
One reason would be so the dev can debug and see how it arrived at its conclusion. You can actually scroll through the whole timeline of it's workspace while it works. Pretty cool.
to play devil's advocate, maybe so it can follow tutorials written for humans?
@@PepperoniPapayayou are right. If it worked as copilot but for all of the codebase you couldn't understand a thing
You realize Devin doesn't have eyes, right? The web page is downloaded, and since it's already a web page, and you're using Devin in a browser... Showing it would be trivial.
The browser, terminal, etc. mainly serve as visual aid for users and VC so it seems cooler
Try curl-ing a JS site
I’m a novice developer and whenever I’ve tried to use ai for something that had no documentation online the ai would be so full of mistakes it was unusable
learn prompt engineering
@@btm1 No level of prompt engineering can write you a device driver. Coding will awlays be a thing.
@@rusmaakatupal4723 sure, just like agriculture is still a thing...but that does not mean there won't be lots of jobs lost in the process
I've even seen it make up nonsense about well documented APIs that aren't that extensively used.
You need to change the scope of your question, and reiterate. You will learn to make good use of the LLM. It actually takes practice and getting used to. It will work with even unknown concepts as it is capable of reasoning and following the pattern you provide.
Some of the big mistakes people make using LLM: People give up too fast on the LLM and avoid iterating, and people also ask the wrong questions within the wrong scope, and do not give enough examples. Eventually you will learn and know exactly which questions the LLM will most likely be able to answer, and you can get to those very questions using the LLM as well. It's easy tool to underestimate because we are used to work in a certain worflow from years of experience without LLMs.
I've been looking for a job for almost 2 years and I have been ghosted towards the final interview stages for positions I was sure to get probably because of my time zone and location. After seeing Devin on Twitter, I closed my PC. I even took up a CS degree last year to get an upper hand and take time to get an understanding of concepts that other senior devs kept mentioning like memory management and so on(I could have learnt these solo but I still needed the CS degree to back me up). My productivity over time has been decreasing logarithmically. I feel cooked at this point and even if I was considering a career change, I put so much time into what I actually loved to do. Devin is awesome and helps speed up development but on the other hand, it reduces the chances of someone like me with so many disadvantages to make a living out of the career which stings. Just wanted to put that out, I'll get back to my grind and hopefully, things work out. I'm currently working on a project to solve a problem in my country with other devs in the same state that I believe would get me an upper hand, hopefully.
good luck bro
Good luck man, its hard to get a job nowadays even for experienced software engineers
Stay strong brother, I believe in you and trust me you will get a job and things will work out for you !!❤️💪
we all are fucked anyway! just go all balls in, what else can you do
i feel like a cs degree is basically the bare minimum at this point, the biggest things that make new devs stand out are good projects and internships
Programmers to Artist : You cant stop progress
Devin to Programmers : You cant stop progress
Programmers never told that. More like you are not required.
@@dvoiceotruthartists are required to fix the shit ai makes. There are literal jobs for that. Its the same shit.
poetic 🧑🍳👌
Listening to this while coding, concerned that I might be coding for the last times in history cause it's going away, and your soothing tone is just what I needed to be able to keep going (even if you end up being wrong). Thanks
1:10 This is a really fair take. Im currently going through Uni, studying CS, so all I am looking for is to learn, and not necessarily catch up with whatever level AI is currently at. I do not have 5+ years of experience, and yet AI seems to make me feel like thats what I need just to get started, else Im good for nothing. You shouldn’t need to be an expert to simply be able to contribute.
In my opinion, unless you need structured learning, college is a waste of energy and often times a major financial fuckup unless you have the government paying. As for you getting work as a SWE, focus on what you would enjoy building way more than the money. Don't be the person who becomes a lawyer to have a rich lifestyle.
As for your experience, at the end of the day, business people value results. If you can deliver and solve a problem and give value, and they are not complete cheap asses, broke, or exploiting you, they will pay.
Im a 15+ year dev who started as a pre-teen, with many hats, and I have 0 concerns about this and use AI near daily. Though im also doing very unique stuff and so another business lesson is specialize. The easier something is to do, unless there is another market factor, the more the price will be pressured down and commoditized. A good niche where you can deliver value will earn you a living.
@@scooterman30 Im in the same boat im about to graduate with masters in comp sci next year. Got decent scholarship, parents contributed some part, rest paid from part time jobs and was kinda scared with all these new AI coders popping up every other month. I even started learning aws cloud because of this and cleared foundation level cert exam. I was building a full-stack project and saw this AI coding stuff, got demotivated and thought is there even any point to build my own project when you can do it with AI but your comment gives me hope. Thanks
Focus on understanding the theory, the fundamentals. That’s the whole point of getting the degree anyway, and luckily, that’s not something that can be automated by software anyway!
@@spralwers theory is just another way of saying academia imho, and even that can be learned free if you really want to. I care more about practical and getting problems solved.
@@scooterman30 what do you specialize in? Im trying to find something I like but nothing is really clicking.
From my experience; in recent years it is more of an issue of adapting to ecosystem of companies I am working at than doing actual coding. I have no issue to code even very complex code that is required; but it is frustrating to get all the required info from management and/or mark all the "checkboxes" of workflow.
Right, and if it'll get exponentially better at solving ever-increasingly difficult programming problems, occupational competence will be decreasingly dependant about pure coding ability - the thing I feel mostly differentiates juniors from seniors? How is it that it's the well paid senior devs at highly paying positions that feel smug and irreplaceable?
If you fear about being replaced:
The “I” on LLM stands for intelligence . If you train a LLM with data that says 1+1 is 3, it will always says that 1+1 is 3. It will never have a critical thinking and actual learn and think. Its just a word calculator. That’s all it is.
I'm not a software engineer, can you explain it a little bit for me? Any reply will be appreciated
@@valerioharvey7289 Do you know the dictionaries where your phone learns the words that you use frequently and next time you write a similar sentence it suggests the words you already used? It's the same with LLMs but instead of short messages, they put it through thousands of internet articles etc.
@@valerioharvey7289 AI does not have the ability to discern lies or come up with ideas on its own, it simply gathers information and presents it to the user. The "AI" isn't smart enough to reject human bias.
Tbf if all our sensible experience of the world was 1+1=3 we would though that too
@@user-og6hl6lv7p Buuut then you realize AI has the ability to discern that 1 is a single unit, and that 3 is three of those single units, therefore that 1+1 couldn't possibly ever be 3.
I feel like passing technical interviews is not surprising at all, it’s right up the alley of an AI, single problems that have been well documented and have plenty of examples on the internet, from what I have heard, most AIs (including Devin) eat shit when they have to code at scale for a large code base, some even claiming it has bugged out and deleted working code after trying to solve a bug in it. Not even to mention how, since AIs like to use boiler plate code off the internet, they tend to introduce a lot of security vulnerabilities. There are plenty of other things too that I’m too tired to think of.
Yeah exactly. I'm still waiting to see a workflow of one of these AI devs that's on an actual large real-world code base and not writing some simple code snippets in a vacuum. Then see it have to figure out where to put its new code in that existing code base, and what code that already exists to remove. It's a similar problem to self-driving cars, a person always has to be watching to make sure the thing doesn't drive into oncoming traffic.
Ah yes, can't wait for everyone to replace Juniors with AI and then figure out in 10+ years that there's no more Juniors to fill in the blanks of Seniors and AI Garbage (If AI doesn't receive info from humans it tends to bloat itself, and if it doesn't receive NEW content it won't grow and won't be able to deal with newer frameworks)
Where did all the Juniors go? I thought we had a line of them lined up?
Yeah, that was 5 years ago when people figured out they wouldn't be getting a job anytime soon because you went replacing everyone with AI and now everyone went into trades or manual labor.
Edit: And yes, I've watched the video and agree on the fact that AI isn't going to replace jobs, but its not the experts or actual workers these AI companies have to convince, it's the managers and sales people who have no idea what the hell is going on and just want a bigger profit for themselves.
I mean in 10+ years ai can get to insane levels, but then any job would e replaced so it wouldn't be just our problem anymore😂
a) if ai keeps improving solely off the back of more data and not better algos - it's already doomed
b) I understand the sentiment, but really? Trades and manual labour? Are there no other worthwhile majors?
As a former electrician it's kind of humorous every time I see software devs talk about switching to trades (seriously or ironically). Most people won't last a year as an apprentice, much less makes a career out of it just to get a salary of a janitor in return for busted knees and back by 40. Not the main point but compelled to comment
@@markerickson6320 Hey man I agree with point a) but it just seems that these companies just release a new version of an LLM and market it in some different way, I'm pretty sure there's progress behind closed doors, but without concrete evidence it just seems that it's just an LLM gold rush.
About point b) and trades and manual labour, trades only take 2 years to learn for, considering a lot of people fear that AI will take their job, then time is on the short side and you want to go somewhere where AI can't automate, sure Surgeons, Doctors, Electronic Engineers, and others are better long term, but if people are going to get fired in mass even if they went to stuff like Computer Science becaues AI does it better, then you can't really go back to University for another 4 years or more if AI took your job, by the time you finish those degrees AI would've affected them as well. Community College is also a good option because it's faster, and time is the limit because if layoffs do happen, not everyone is going to have the luxury to move back with their parents.
I understand that it's not the best, but its what I would personally do if I were caught in that situation.
@@aarholodian And as an electrician you've got one of the less back breaking jobs in the trades. I did months of work experience with my dad who was a carpenter and joiner, that shit is hard. Mega respect for tradies. I've watched carpetners singlehandled monkey climb up half built house frames and hold trusses in place with one hand and using a nailgun with the other, genuinely most office working people could never last a week in that kind of job.
I wish Devin good luck trying to decipher the lack of requirements in most companies
😂😂😂
The vague requirement will keep Devin spining in the void. Poor Devin. Poor Karen also BTW...
Poor devin 😂😂😂
Yeah, recently I was implementing some feature and most of my time I was not coding but investigating and refining requirements
You could argue one developer could discuss the requirements and Devin would do the coding. With the speed an AI codes this could replace thousands of developers.
That said I don't think AI will ever be relevant as a developer unless we create an AI that thinks as a human and understands context and if that does happen then we don't have to care about jobs anyways since AI could supply us with all we need.
Most useful thing in the first demo is that initial plan it creates and collecting the documentation. I approaches something i would pay for because that can really save some engineering time.
Devin saves so much time that Cognition has its team putting together demos at one in the morning.
i hate how LLMs are put in the work we like doing(and as devs the code is the thing we are good at not the rest), just put LLMs in the tech stuff we dont like doing. Make it write documentation for an existing legacy system, make it prepare a cool presentation for the executives for the last sprint completed, make it flag commit messages when they are not descriptive enough(or somehow enhance bad commit messages). Make it look-up obscure bugs that might be found in code submitted for a review.
Devin is probably better at creating his daily post on LinkedIn to farm impressions for his company.
Ha, Devin read documentation… because documentation is always up to date 😂
True Rust devs don't need docs, this brings the raw-dogging experience of chaos and exploration.
Just run Devin on their code and make your own internal docs for external code
(totally can’t go wrong)
devin can read the source code too.
Even if Devin could do nothing else but read the documentation, read the code and submit pull requests for fixes to documentation, it would be great. Assuming it's cheap enough to use. The demo to read a blog post and repeat the same thing with ControlNet for about $750 is not a threat to any human developer.
That sympy bug was already fixed in 2019. Devin was run on an old fork of the main sympy repo. Issue #17148
they literally tried to fool us 💀
Solving like 10% of the issues seems impressive until you realize it's the easiest 10%
Thats true but you need to consider the fact that problems are not uniformly distributed. Most of the SWE problems that businesses have are easy. Some businesses use only spreadsheets, forms and other already existing tools. Some need small customizations. Very hard problems occur very rarely. It still means that Devin might affect thousands for SWE job positions.
@@darekmistrz4364 yeah that's very much true but I do think low level developer are worthy of replacing, im fact I think they actually should be in my opinion because of the learn software earn money society around software engineering this field is full with developer that earns a lot without actually putting any value to the company, which I think they should be replaced.
I spend time every week testing AI tools to see if any can improve my workflow.
The most productive tool for my case is LM Studio running Mixtral 8x7B Instruct.
It's a brilliant example of how shit the benchmarks are or how much overfit there is towards the benchmark.
I ask it to write a function that I've written a stub for while I keep going. When I'm done with my task I check if it's function runs or if it has to to re-generate.
My 4090 goes hard and every 8 times it produces a junior level function.
Edit: Worth mentioning that I work in the company that I co-founded and own more than a third of so my incentives is maximum productivity.
That video is such a marketing tool. All of this just reeks of a startup that really wants to get some kind of market share. It's crazy how much of this comes from the crypto playbook. I've been in the industry for 25 years, I'm not worried about Devin taking my job anytime soon.
There's new hype every few years and it rarely materializes. I was there for Enterprise Java Beans, SOAP and a few other "life-changing" silver bullets. Just like Prime said, this is just to get money from investors.
@@ArturdeSousaRocha yup, exactly. That said, even comparing to how people marketed things back in the java days, it was less insidious compared to how. With crypto, so many people just stopped caring about pretending to have any ethics.
With such vast experience, did you predicted SORA or GPT-4? Nobody has a crystall ball to see the future. Don't be naive, stay prepared.
@@btm1 True, every once in a while there comes a disruptive solution that upends the status quo. Usually, though, it's not accompanied by hype. Things like WWW or social media just slowly gained traction and only *then* businesses started piggybacking some hype on them.
@@btm1 not being naive, I'm being realistic. I've worked on AI myself and I've read many of the research papers, there is a lot of evidence to point to the fact that we are not going to hit a point where these models are going to exponentially get better. Probably the biggest thing is that they're going to run out of usable data by 2025 which means that for the models to get better, they will need to improve the algorithms which I'm sure will happen but not at the pace they have been recently. There was a research paper on co-pilot, where ever since it's inception, code quality in general has gone down. This makes sense because these models don't really select for quality code as much as they exist to just solve the problem by any means possible. Solving a problem and creating something that can be maintained for a long period of time are two different and distinct issues at the end of the day. For non trivial usecases, the code needs to be maintainable or else even these AI workers will succumb to technical debt.
To me the most compelling piece of evidence are the numerous examples that show that these models don't really understand anything about the training data. Take for example the recent snafu with Gemini where it declared c++ unsafe for users under the age of 18. The fact that it couldn't figure out the context of the term unsafe with regards to programming versus unsafe in a general sense proves that it doesn't understand language as much as it just predicts patterns. Even prior to this Gemini incident, there was a paper on alphago where the researchers showed that they could consistently beat the algorithm using one of the most foundational concepts of the game. If the model truly understood the game of go, then it wouldn't have been susceptible to such tactics.
actually just wanna say thanks for reassuring words
^^
Did Primeagen predicted SORA or GPT-4? He may say some things that are true in the present, but he has no fucking clue how much will AI evolve in the next few years.
Hell, even the experts were surprised by how much progress was done in the last years and the emerging properties that appeared in AI systems.
@@btm1 prime's reassurance doesn't come from the fact whether AI will advance or not.
even if it does advance people won't lose their jobs
@@huge_letters my bad, I guess technological unemployment is an imaginary concept
@@btm1 i'm sorry you're so insecure about your pov you need to crawl comments in search of people being hopeful about future of software of development to tell them how wrong they are
15:15 I'm extremely curious too to see how that turns out.
I was able to be in a flow state for a long time working on a non-programming side project where I mainly used chatGPT and google sheets, but chatGPT wasn't there as a solver but as an unblocker of sorts. This was something creative and while AI can't give you anything particularly creative, seeing something *wrong* makes it easier to figure out what you actually think is *right* when you feel stuck or out of inspiration.
I have awful writers block, but love editing. llms are a godsend for me creatively
@@IsaacSchubert I love both, but when I'm stuck on one side LLMs vhanges the one way path into a complete road.
Usually you can only switch from editing to writing, LLMs unlock the other direction.
“Even if I knew that tomorrow the world would go to pieces, I would still plant my apple tree.”
~ Martin Luther
Implementing a solution is just a part of my job. I spend a lot of my time working through design, pushing back on managers, and communicating with stakeholders. Obviously coding is important, but being in SWE is so much more than that.
If anything, I'm hoping AI will save on engineering time and allow me to focus on the human and design-facing elements of work. Thank you for this refreshing take, I'm so tired of all the AI grifters and fear-mongering online recently.
As an developer with 4 years of experience, I am extremely worried. The market is such a mess that I can barely get an interview and when I do, and I do well on the interview, they tell me that I have been outperformed by someone with WAY more experience than me.
it took me a while to realize how over-hyped the generative-text goldrush is. the main problem I see is that it completely de-emphasized reinforcement learning and other forms of training/pattern recognition. as a result, AI research and funding is actually slowed while a pointless joke technology receives 95% of investment
VC money must flow. 😂
absolutely loved this video. thanks for making me feel better ❤
I understand the cope but the truth is he can't predict how AI will evolve in the next years. Stay prepared.
Thanks man, I was starting to panic. And I've been developing software since PHP didn't have classes.
Finally, a super based take on all the AI fear mongering to attract venture capital. Really appreciate this video!
It is the first and only video I have seen on RUclips which gives a realistic perspective on the AI gold digging.
There’s nothing that’s ever had exponential growth. At best, it’s logistic which eventually plateaus.
sometimes it's arctangent
We really are in the end-times
Devin (as of today) is nothing more than a few tweets and some YT videos. There is no way to actually use it.
It's 2024 - this is considered a groundbreaking achievement in software engineering. Some tweets, a promotional video, and a blog. That's it ! Let that sink in.
I think I will wait for having at least half an hour of using it before forming an opinion.
im switching to electrical engineering man im tired
it replace with ai too
@@werren894nah EE is the fundamentals of ai
@@werren894not really
@@sid4579 what a clueless person
@@sid4579 we have better infrasructure or machine than you doing soldiering, and a lot of simulation software to simulate semiconductor, i would say any engineering job is AI anyway human are just the front end you have no control over science anymore.
This feels like an NFT/crypto 2.0 all over again
Hello to anyone reading this. I'm currently on the verge of finishing a high school with an emphasis on computer science (I'm from Europe). I've participated at coding Olympiads, finished CS50 and The Odin Project, and dabbled in game development.
A couple of years ago, if anyone were to ask me what I'm planning to pursue in the future, I would have probably said some kind of software engineering job. However, considering the current state we are in, I'm not so sure I have the security of knowing that my skills will even be relevant anymore in 4 years.
I've been thinking about trying to put less of an emphasis on coding, to explore other parts of computer science, such as cybersecurity or embedded systems. Maybe by the time I finish university those markets would not be as saturated as a frontend developer for example.
I would love to hear the thoughts of anyone with more experience, as I feel quite lost in this field at the moment. Thank you for spending the time to read this!
The discouraging thing is that it's really hard to predict what AI will actually be able to accomplish in a few years. It could end up being a strong booster for many people or it might be wholly snatching up jobs we don't even consider now. If it's the latter though, we will have societal problems on a scale that they can't be solved by individuals.. and at that point we need big changes in the society or we will be heading to a revolt.
read the unplugged alpha by Richard cooper
The irony is that Cognition, makers of Devin, is currently hiring SWEs. Can’t they use Devin?
"AI will probably be smarter than any single human next year" has the same energy as "In ~2 years [~2018], summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"
congrats, you just found out nobody has a crystal ball into the future
@btm1 we get it, you want to suck Elon off
Some people think that fast progress is equivalent to exponential. We could definitely end up in a Moore's law scenario where we have rapid progress that eventually stalls out due to technical limitations or whatever.
Super wishful thinking, public AGI in months to 3 years max, ASI one year after that, Sam Altman will get trillions for a system that gets us to ASI and AGI achieved internally already at 99% probability. Trying to slowly drip technology to not induce mass panic , but eventually will be forced to let some beast out (even it its tamed down and not AGI will be societal changing) . Hardware won't be a limiting factor for very long. Groq chips are better than Nvidias, Quantum Computers are around the corner, many new technologies are on the brink of being released , and Devin is dumb compared to what exists internally. Even Claude3 seems smarter than Devin.
The technology we see is in the infancy stage, it'll explode within 2 years and be fully matured. Thinking linearly will hurt you more nowadays.
We can't limit ASI , if it's really millions or billions of times smarter than us, it will just create new hardware or technology that will be exponentially better each time, and even if there's a limit on it , it would be further than humans can comprehend. They will be able to bend physics to their will and master all the laws of the universe. Discover things humans would never discover too.
We just are looking at hardware and software, maybe there's flexiware/middleware/physioware/bioware/ or something inbetween or in other dimensions that we can't perceive.
Start of AI history, human's final invention.
@@phen-themoogle7651bot
@@phen-themoogle7651quantum computers are not around the corner, this would require an experimental (and probably theoretical) solution to quantum decoherence
@phen-themoogle7651
> Super wishful thinking
You described your post accurately, you must be some sort of AI or something.
@@phen-themoogle7651where is your evidence for any of this? There is no indication that any sort of agi is possible much less achieved.
Devin, the next first AI software engineer.
imagine jail breaking an AI interviewer and catching the CEOs position :-D
All the devs coping hard… it’s over by Devin 3
What happens when we adopt a new framework/way of working and there's no training data???
You can basically already see the effects of this with e.g. ChatGPT. Try to get it to write a working Minecraft Datapack. It can't, because the available documentation is not only horrible but also out of date.
You nailed it in the beginning, the mental gymnastics to call this a "first" are impressive.
been a software engineer for 8 years, i use AI tools all the time. you are correct, it isn't taking SWE jobs anytime soon. ill be worried when an AI can do my job, but right now none can do that. in fact i have to struggle with current AI capabilities to keep on accelerating my current workflow. so I agree with you that it has to 100x the amount of capability it has right now to even become close to replacing jobs
If you decide to take 1 year off, do you think you will easily find a job that pays you the same amount when you come back?
I guess it can in around 3 years
Nobody has a crystall ball to see the future. Don't be naive, stay prepared or you'll be caught with your pants down.
"There will be fully self-driving cars next year" - Elon Musk, 2014
Job vancancy: Devin developer.
10 years experience in Devin.
Agreed, I like AI, but I certainly see all its pitfalls. My stance is, as long as it's based on 0s and 1s, we aren't getting generalised AI/advanced AI.
I think its conceivable we will engineer an AGI that behaves like a brain running on modern machine architecture, but it won't happen through LLMs.
Imagine Devin replacing the whole team who actually made Devin xD
😂😂
I have to watch Prime on 2x speed, which is hard bcs he talks so fast but man... there is so much mumbling and bullshitting around important, genuine good takes
Zoomer attention span
But thats why we watch him. You wanna cut away the fat from the bacon?
Stupid comment
I think this is actually a good thing because if you want to stay in this industry you have to continuously working and learning and try to improve, unless some company makes an "magical ai" that is able to replace all the software engineers then I might problem but the nature of "if you don't work hard you will be replaced" isn't actually that bad. actually its good to see front end dev getting replaced.
Did some research looking into the team at Cognition Labs and its basically all mathematics and iOi Olympiad gold medalists and codeforces grandmasters. LOL these people are literally at the top of the top of their game in Olympics mathematics and competitive programming. I do believe they will achieve their goals with Devin. Give them 1 or 2 more years and it may shock people, specially the ones in denial right now.
Don't fear AI fear MBAs
What is wrong with macbook air.
@@kiattim2100🤣🤣🤣
Where is my self driving car?
2 years, max 3, quote me on that
.
@@MrMCKlebeband Its a moving goal post, just like fusion power
@@MrMCKlebeband3 years later.... “just 3 more years quotes me on that”
its already 10 years and google was on the driver seat. Still we have nothing
@@darkbird2724 soooo close, but still not there. It seems that the final self driving is 70% finished but the last 30% seem to be extremely difficult. Maybe Devin can ... or not?
We need to talk about Devin
I am so surprised that there are people out there who are still surprised by this as if no one knew that it's already available, and that it will be updated in two months that it will just be impossible to make flaws. It's such old news....
This guy must not have room in his fridge for food when it's filled with so many bottles of the COPIUM he's gotta be drinking too not see where this is leading 😂😂😂
I really liked the take of "stop selling AI to VCs and start selling it to people". W take
He only said some things that are true in the present. Nobody has a crystall ball to see the future. Don't be naive, stay prepared.
@@BeerMoneyforTokyo yeah but I think what triggers me the most, and I think that for Prime too, is these companies putting out these bold statements to get VC money without thinking one moment about the consequences of their claims on the rest of society. Specially when the claims are highly dubious, like these
That point about flow state is very very true. I’m at that 6 years of experience point now so I’m not scared too much by these AIs and often I find myself using gpt more as a colleague whom I talk about an issue with, rather than actually copy pasting actual code.
Same. I use ChatGPT to help me code everyday, but I don't think I've ever once copy-pasted any code directly from the prompt outputs. I use it more as a conceptual research assistant and glorified search engine than anything else. If I do have it write code, it's so that I can better understand the language/framework I'm working with, and then I'll actually write it myself for the project I'm working on.
tl;dr it's a programming assistant, not a coding assistant. Works best if you already know how to code. Important distinction right there.
Yeah I don't like his coding style 😂 but he can be helpful explaining (when he's not completely wrong or on a different topic)
@@sedmidivkahe ?? It’s not a person, call it ‘it’, before society falls apart
That copilot suggestion lmao
The value is the way it plans steps and applies changes to existing applications. The actual value is not only in the code it generates, but the automation of steps which can be applied in more use cases. Perhaps in the future you can prompt it something, then it can generate the automated steps to achieve a task and do the task. For example, you can ask it to integrate with a smart home service it has never integrated with, it can automate the integration by creating the code and then perform the task. That would extend it's capabilities beyond code generation.
I'm new to programming just like a lot of people and I'm glad him and other youtubers are addressing this and making sure newbies know its ok to keep learning and not get discouraged. F**k you Devin
It only solves 14% unassisted... we're still a ways away from this being a concern...
on the other hand it went from 2% to 14% in how many years? 2?
5 years ago it was at 0-1%, 5 years from now it will be higher. Not being a doomsayer, just sharing my observations that technology _always_ progresses until we all die.
It's false logic to assume the curve is linear. It almost never is.@@Rexhunterj
They are solving the easy part of a hard problem, so their progress looks better than it really is.
The other big problems come with questions like "What happens when it fails?" Does it just put in a non-functional best guess garbage code into your codebase? A big issue with ChatGPT is it rarely says "I can't do that", it just makes up some crap that looks like a solution but fails. Imagine a codebase polluted with nonfunctional or buggy code like that.
Great it may do a good job 14% of the time, but who is cleaning up its failures?
I think the prime is missing this point; the way he looking at it is like, the day we reach full llm automation, is when SE start losing jobs.
I think he is ignoring the fact that LLM/AI just increases the competition in the market, and that how SE will lose their roles.
Right. I see a lot of copium here, but Big Tech has already made their stance clear on this issue with layoffs. Even if AI can't replace programmers yet, companies are getting leaner to invest in AI and are actively looking to layoff where they can.
It's not a percentage thing. AI can't and likely wont be able to do what even junior devs can do, because it's not a matter of being able to regurgitate variations of pre-existing solutions, it's being able to logically understand WHY you're writing the code you're writing.
All that will happen is in 5 years, companies that rely fully on AI will be screwed and the companies that hired and developed people will succeed.
The entire industry isn't going to collectively decide to stop hiring all at once.
Now developers know how artists feel…
Nope. This demo made me even more confident that these robots are not taking our jobs.
@@DeveStarr You dont understand, nobody buys AI generated stuff. Ever seen how people flip their shit about asset swaps in games made with Unity? People dont want AI synthesized art or voices in their games, the mindset is "why do have to pay 70 dollars for something that cost nothing to produce?"
Just pointing out that first it was artists being told they’d be replaced and developers would never be replaced.
@@DeveStarr you have a crystall ball into the future? Who predicted SORA?
Excellent rant. This is the same rant I had in my head. Thanks for getting it out there.
If you had 10,000 monkeys typing 10,000 years you could also come up Devin's non-trivial output.
Im a first year college software engineer student, and this whole devin thing is making me very anxious lol. I dont want to quit and stop learning, so what kind of programming thing should I try to study now? how can I find the most important area to go into?
I was thinking of stuyding how to work along with these new AI's but how can I look for that to study lol?
Go and work as a miner or something
Devin is actually controlling the company, hiring cringe people to do "believable P. R.", and software engineers to work on what he's supposed to work on, and in the meanwhile he's working on his own code.
2019: learn to code
2024: learn to cope
Software is about to reach new heights of shittiness. Bugs. Inefficiencies. Bad ideas that never should have seen the light of day.
"All issues are solved by human pull requests." says the robotic lizard