Short answer yes, Devin is a scam. Theo did a really good job covering and breaking down the Devin promo announcement a month or so ago. Devin is a glorified API wrapper.
There's a reason there's not even a whiff of architecture anywhere on the internet - bc there really isn't one. They don't specify what degree AI contributed to it's functionality. It's an API wrapper, some hard coded scheduling scripts etc
@@Nootlink the primary issue is hype-traning something as groundbreaking, new, incredible etc. The naming "the first AI programmer/engineer" implies much in the way that it clearly is not
@@Nootlink you’re not wrong. I was however referring more so to the point that Theo made that if they had at least built their very own ai model from scratch, it would be one thing and at least be something different.
It's not just Devin there's a surge of tech 'startups' without any product but a huge marketing budget. Rabbit is another example. Both of these are wrappers around Open API endpoints without anything to sell.
Exactly. I (mistakenly) thought, that unlike 'crypto' and 'web3' that with AI any companies or startups we'd see would produce something of value or an actual business. We luckily are seeing stuff of value but for every interesting offering there's 3 garbage 'AI' offering. Increasingly becoming a market similar to the metaverse fiasco.
Getting highly ranked on leetcode requires techniques that are counterproductive in production. There is a big difference between coding as a sport and coding professionally.
Also, just because someone truly is really good at doing something doesn't mean that they'll actually do that thing properly if they don't need to. A fairly classic example is the number of top level speedrunners who have cheated runs, even if they were skilled enough to have conceivably done that run legit. The Cognition AI team could be the best developers in existence. That still doesn't mean that they actually bothered to make what they said they've made.
Top at leetcode for these bots means nothning lmao. Anyone could google the answers/algorithms required. This bot is just an enhanced search, been that way since the start.
I mean, they definitely have potential to become really good. But they need to get actual work experience and do development to become someone useful. And to be a really good software engineer, you need to hone a lot of skills other than raw coding.
Wait, you mean like Dollhouse? Like, in the dreaded "The Attic" that's used for dolls who don't comply? Sounds about right. That's why Elon is so big on Neuralink.
IPUs A bunch of Indians in a building powering the latest rigs remotely and helping you with all your computational needs. Until India's populatiom starts to decrease, which it eventually will in some decades. Then we gotta find a better procressing unit.
There is a great video by Internet of Bugs that you should check out where he shows HOW BAD Devin is - he even demonstrates how it's done correctly (with examples).
@@reanukeaves I mean, it matters in the sense that all the skilled people can learn the truth and move on, while the dumb investors can lose their money and get stuck holding the bag.
@@4m470I wish. It's rarely the investors that get left holding the bag or losing their money. VCs are *REALLY* good at hyping things up enough they can find suckers to sell their shares off to before the truth comes out.
Devin is completely a scam. Saw a video a couple days ago where a guy exposed how the promo video about Devin supposedly solving an upwork task was fraudulent, with Devin making a ton of errors like referencing non-existent files that it then fixed. Also, the timestamps indicate the relatively simple task took Devin about six hours. It's a scam meant to raise more VC money.
Devin is based on GPT4 so it'll run into GPT4 problems with hallucination the more context you try and fit into it. The "advancements in long term reasoning" is complete bs. There is no long term reasoning past the context window. The most you can do is store previous text in to a DB with some sentiment about what its about and then reuploading that part of the text into the the context window when Devin thinks it needs it. But this won't happen often because of the nature of how the GPT4 context window is limited in the first place
I don't get this new wave of "AI startups" where it's just some interface on top of chatGPT api. This might be the "thinest" layer of a "product" I've ever seen. I'm not saying there's no need for products like this, but how the hell is that a full startup company with investors? It's literarily hackathon level projects.
What if the video is copium and Devin isn’t a scam? What are the odds that all these rare gold medalists who work on the project are scammers? Every single one?
@@numalesoybea1348 its bc if you look at the climate, context and evidence, it shows that the Devin team are pulling off a bunch of smoke and mirrors. They're not leading with truth. On top of that, all the other well funded tech companies like Google and Amazon have gotten caught faking their AI products. You think a few unproven CS guys are suddenly going to change the world building a UI around someone else's product, which is black boxed? C'mon dude. It could happen, but the climate, context and evidence isn't strong.
@@numalesoybea1348 very, very high? Like, pretty close to 1. "Rare gold medalists"... I mean, I dunno how high I'd rank on leetcode, because I'm too busy producing actual work to write leetcode. Same for all the best developers I ever worked with. Ranking high on leetcode proves absolutely nothing about someone being a "rare gold medalist" except that they've got a lot of spare time. You're just arguing from authority, when nobody had ever heard of these people five minutes ago so they're not even authorities.
99% of the work of a project is the infrastructure around the project. Security, metrics, logging, error handling, outage handing, scale, multi region, deployments, backups, dashboards, support tools, support infrastructure, incident management, documentation. The difference between releasing a product and releasing an enterprise product.
finish product is 100% ready sell to customer; demo is 20% ready; prototype is 5% ready; devin show it core function, it replace programmer = 0.1% ready fund me or programmer die!
I'm just glad that soon we won't have to do any more more tedious tasks like coding and we can get back to focusing on what's important, like consooming products and scrolling social media. 👍
and how are we gonna consoom anything if no one has enough money to afford anything. seems like those in charge corporations are not really smart themselves
If you haven't seen it, I really recommend Internet of Bug's video 'Debunking Devin: "First AI Software Engineer" Upwork lie exposed!'. He points out that Devin fails to complete an upwork task, despite this being part of their marketing pitch.
I mean it's kind of good news in here for actual human programmers in the sense that the snake oil web3 schmucks are likely going to screw up AI coding to the point that the industry takes at least an extra 10 years to develop.
There are strong reasons to say it won't develop. Everyone's so hyped but the reality. Most of what AI is actually doing is mildly improving autosuggest, or giving anime nerds underage porn I mean ah... Wait no it's serious business stuff I swear! My 12TB of "anime sexy 3yo in diapers naked" prompts is totally development work!
@@KevinJDildonik This, unless you're in the business intelligent side of things, computer vision or image generation/ text generation, no matter how much better AI gets, it's always pattern recognition at it's base and won't be anything else.
@@mortvald very, very, very, very, very, very brave of you to say AI will never be anything other than pattern recognition. "Never" is a VERY long time. Sounds a lot like the apocryphal "no-one will ever need more than 640 KB of RAM".
@@BittermanAndy Nah, this is my field, been my field for 15 years and I can confidently say it. I can recommend anyone who don't know the first thing about AI to look up a video about regression algorithm. It's simple and easy math and does show how prediction models work. once you see it you'll immediately see the flaw baked in. AI is a tool for specific tasks, been here for a long time and all that happened lately is that big data enabled more complex models but they're essentially the same. AI hallucinates a lot and will always have that 10% to 5% error margin (and this is for world class AI models) it can never be employed in engineering where precision and determinism is everything. for art sure, a mistake on a draw or bad prompt won't do much harm, for translation sure, a few mistakes won't harm anyone. but for anything that requires precision over prediction? AI will never be the answer AI was born just like computers were, an attempt to mimic the human mind, now i'm not gonna say we're not gonna pull it off one day, we still have a long way to go and need to understand our mind inner working first but even when we succeed, it'll inherit the flaws of human mind. We're good at pattern recognition sure, but we also fuck up because of it.
How do you define the term "scam" so that Devin isn't a scam? They're trying to sell you something that not only hasn't been built, they don't even know if it can be built.
@@spicybaguette7706 they ain't stealing anything, the VCs decide to hand them over their money willfully, it's not their fault that VCs pump money into things they don't understand, that's a flaw of the system, not theirs.
My 2 cents honestly, the whole Gen AI bubble seems to have been pumped by the wet dreams of corporations having workers fast replaced and splurge into the immense fast returns without having to pay salaries & benefits.
To be fair, anyone claiming authentication is easy to get right doesn't actually know what they are talking about. It's easy to get something that seems to work, but much harder to get something that is actually secure. Of course there are services that abstract the problem away but that's basically the point. I wouldn't trust a junior dev to get it right, let alone one that is prone to hallucinate.
@@3_smh_3and even implementing OAuth correctly is difficult😅. And OAuth is not really created to simplify auth, it's to provide a universal interface for authentication providers
I started out with rust (and programming in general) when chatgpt came about. The rust compiler seems to spit out rather detailed info about the bugs. So in some late night sessions, i found myself just copy pasting the compiler output back to the gippity. It was a simple pingpong process for simple solutions that i didnt know how to handle myself at that time. When things got frustrating, i sometimes thought of a simple program to to the copypasta pingpong for me. Maybe get the gippity to write that for me. Just watch the chaos unfold instead of actively clicking for it to happen. Even in those moments late at night with some "why doesnt this f'ing work at all??" mentality in the head, and being too frustrated to do the research myself, i was always sure that this would not lead to the code i wanted to write since the gippity was only a crutch telling me bits of syntax that i needed to get going. Letting it take the wheel even from the very beginning for me, always led to worsening outcomes, deadlocks, version conflicts and unneccessary rewrites
Funny thing: On their website, at "come work with us" link, you can see they are hiring (human) Software Engineers :D I know it makes sense, but it's still funny...
if this is in fact the result of VC funding, VC investors seriously need to learn the space. So many of these (maybe all so far?) AI startups are clearly pump and dumps. If I could short these companies, I'd been be shorting every last one of them. They stink to high heaven.
My take on this is as follows: If I was a legendary grandmaster CEO, I would not personally write the front end for the company. If my main product was the AI I would dedicate all my best programmers to that task and hire out the job of creating front ends. The problem here though is they claim to have created an AI of the very sort that you would want to hire out such jobs to, and it really does beg the question as to why Devin did not write the front end apps and if he did, why are they so bad.
Good luck maintaining whatever crap AI builds. Bitcoin is overly complex and inefficient but atleast it has purpose - druglords can transfer money somewhat anonymously. But this shit is inefficient and has no purpose. You can have best GPT in the world, but if you teach it from all the buggy code which is freely availble it will write even buggier code 😂
Internet of Bugs has a good video going over the specific examples , regardless of their metrics in swe , its the way they are falsely presenting the capabilities .
I dislike the idea of Devin as much as the next guy, but this is an early phase startup building an MVP to do product-market fit. If they are to succeed as a company, it's very likely their product won't be what it is now. Also, the whole "fuck Devin" kickback gets them a bunch of free PR without even having a proper landing page.
Thing is, it's basically impossible to tell the difference between "early phase startup building an MVP to do product-market fit" and "scam", until you actually see the product working. They went public before having a working product. Say no more.
You should definitely watch Internet of Bug's video "Debunking Devin: "First AI Software Engineer" Upwork lie exposed!" which basically exposes Cognition AI for holding back some key information about Devin's real performance, which when examined closely definitely is performing at a way lower threshold than what Cognition AI is making it out to be.
Yep, it is. Hype is dominating the media right now. And lots of people that don't understand the limitations of these technologies, get to believe whatever they are told about it. ChatGPT was the catalyst with its impressive, human like communication
@@redheadedmoos1204 I’m not denying that but that person was being purposefully racist so my response was about them being Asian American from the USA as proven by the competition background.
also 14:08 LOL imagine being in a conf room together with smone, working on a project or smthn, ... and they start out the blue calling rank and hitting u with that goofy ahh title man i would be SHOOK like am i being pranked?? 🤣
There are old scams, aka "fintech" and new scams, aka "ai". Entire post-modern neo-liberalism is a scam. Ever since they dropped the gold standard we're one scam after another scam in this society.
I made a screenshot on the release day of the openings and they were just hiring software-engineers. If they would have found an AI Engineer they would have hired sales people.
A friend of mine had a Professor who said to his class "you better get really good because AI is gonna replace you so fast if you don't" a few years ago (obviously not the exact quote). Devin is the most promising "replacement" and honestly someone with a decent head on their shoulders is gonna be fine. Right now Devin isn't even able to turn a good engineer into a 10x dev, I would concede Devin as a success IF someone who knows the answers to the prompts they are giving him is able to be at least 3-4x more productive because of it, until then id rather have a "go-fer" intern than devin
7:00 the problem is outsiders don't have trained eyes to dicern right or wrong, and many of them blinded by the fear of missing out and greed of being able to do things without learning. Remember Theranos? Any the medical specialists when asked about Theranos said it was bullshiet, but no investor really pay any attentions.
Yup. While DSA is important and a good skill to have, DSA does not necessarily mean being a good engineer and good at building *systems.* I know guys who can solve hard leet code problems with their eyes closed, but can't figure out that they should version their REST APIs and end up changing end points and breaking clients. Or in other words. Being good at designing systems and being good at engineering requires being pretty decent to good with DSA. But being good with DSA does not mean someone is good at building systems or being good at engineering.
This particular "Devin" might be a pile of crap, but "Devins" are inevitable. This is just an iterative LLM using errors as input. Eventually it will be sufficiently fine-tuned to build a viable app from start to finish.
@14:00 regarding no error handling on a project, there is a 3rd option... they did not get payed enough. Like "Go horse" thinking, just get it to work , barely, then when errors pop up fix them ... Defensive programming is a great skill but eats time & budgets, not very well liked by pushy 'investers' or project managers.
Self driving cars is when we can be afraid. If you get real self driving cars it means they solved the inability to generalize. The fact that self driving cars don't work is what makes me kindof think that LLMs aren't enough.
Self-driving cars are especially hard because you need low latency loop. I think you could already do LLM based implementation for self-driving car if you accept is goes about 1 mile per hour to give it enough time to evaluate everything needed and to come up with suitable option. And even then it might need a supercomputer running a single car.
LLMs have nothing, nothing at all, nothing whatsoever to do with self-driving cars. They're not even slightly related. Not even a little bit. (Also, you should get a demo of Tesla FSD. It's not perfect, it's not finished, there are improvements that need to be made, there are situations it can't deal with, but if you think it can simply be dismissed as "don't work" then I don't think you've tried it).
@@BittermanAndy I know that LLMs (with typical text based training) have nothing to do with self-driving cars. However, the algorithm itself is generic enough to train from example driving by humans if you give it enough computing resources. Is it a good algorithm for such a solution? No. But it would be good enough. Human brain works by having a HUGE biological counterpart to LLM technology and it seems to be enough to allow humans to seem intelligent sometimes.
This reminds me a little bit of people selling stock trading courses. If people were so good at trading stocks why do they need a course when they could just be making money. Same situation here. If they really had an amazing AI coder they could open a consulting shop and make $10m in revenue per employee
The problem is developers and software engineers are always self doubting because they are so intelligent. Devin might evolve but it will take a lot of effort and great minds.
Usually ThePrimeagen's takes on things are fairly balanced, would have loved to see some other criticisms of this stuff beyond 'a startup made a heavy webapp'. Typically he'd say something like "I could understand that if your focus is on making a tool, you might not rush to make the web frontend nice' etc, but again calling them amateurs or lazy and literally going through the transpiled JS seems out of character for the channel. Looking forward to the next one!
I like Devin. I always felt programming is just menial task to communicate with computer. With Devin, we may finally dive deep into field of development and creation rather than learning to communicate with computer
Current AI just writes code like the average open source/public repo in github. It even generates the same kind of errors that people make. It has its use, but I still haven't seen anything that really looks like a thinking machine beyond superficial use cases.
Grabbing the money is the goal, the only way to build a team and move fast, pre-product. You don't need a polished website for that, and this is just marketing. Empty shell? Maybe, scam, maybe not.
Main job of AI is to devalue your work only for you to be paid less and work harder. I LOVE SE and i have master degree in it but i am leaving that field and moving to security because i dont want to deal with all that is coming. Gl boys and girls gg and hf.
We can sit and talk about how the current AI tools aren’t great. But the thing is that won’t be the case for long. Especially not with companies and executives laying off engineers and dumping billions into AI. Which is a pretty hostile move towards the workers.
Real talk though: It's legit to use google forms for this type of thing... google has a fleet of security engineers that keep that data from getting stolen. Rolling your own implementation (even if it's just collecting emails in a single column single table database), when the best you can muster is the site that they have --- that's a squeel injection waiting to happen, and that could sink the whole company with that one mistake alone. If there is a breach to google forms, that one company isn't going to be singled out and will be a footnote in a huge list of companies impacted by the breach. And don't think that developersTM won't be pissed off about being rendered obsolete... hacking attempts of all types are very likely to still happen. That said, AI for coding is a complete scam, and has been for roughly a year now. The mess it makes and the lack of basic understanding, makes it worst that worthless, as having nothing at all at least doesn't give you garbage code that was trained on trillions of garbage stack overflow answers. The threat of ai, is going to stifle the amount of programmers learning the craft and I'm fine with that... we're full f-off
There was a nice tool for python developers. Sourcely, it did traditional stuff parsing your code, using some fancy algos to suggest refactorings, and then correctly applying those refactors. It worked very well for what it purposed to do. An honest tool, a bit expensive, but not too much. It was like a linter, but one abstraction step above, operating on the AST, For some random reasons I had stopped using it some time ago, and just forgot about it. Then one month ago I decided to give it another go. Oh man! I hope the hell awaits the product folks from this company! They completely mutilated to product for it to become another "me too!" AI coding assistant. The thing is now completely useless and if you try to use it, create some very short aliases for commands like "git checkout" and "git reset --hard", your fingers and tendons will appreaciate that.
I've been getting scammed ask for something then go and download a model recommended even though it has 180k downloads an then i get spammed with ssh connect attempts. my setup farts at the 25 attempts per second non stop until i do somethnig, but annoying none the less.
Short answer yes, Devin is a scam. Theo did a really good job covering and breaking down the Devin promo announcement a month or so ago. Devin is a glorified API wrapper.
There's a reason there's not even a whiff of architecture anywhere on the internet - bc there really isn't one. They don't specify what degree AI contributed to it's functionality. It's an API wrapper, some hard coded scheduling scripts etc
I mean, everything is just a fancy "API wrapper" if you really think about it
@@Nootlink the primary issue is hype-traning something as groundbreaking, new, incredible etc.
The naming "the first AI programmer/engineer" implies much in the way that it clearly is not
@@Nootlink if you really think about it nothing we do matters because we're all dead in the blip of an eye. Geologically speaking.
@@Nootlink you’re not wrong. I was however referring more so to the point that Theo made that if they had at least built their very own ai model from scratch, it would be one thing and at least be something different.
It's not just Devin there's a surge of tech 'startups' without any product but a huge marketing budget. Rabbit is another example. Both of these are wrappers around Open API endpoints without anything to sell.
Exactly. I (mistakenly) thought, that unlike 'crypto' and 'web3' that with AI any companies or startups we'd see would produce something of value or an actual business. We luckily are seeing stuff of value but for every interesting offering there's 3 garbage 'AI' offering.
Increasingly becoming a market similar to the metaverse fiasco.
@@chuksajeh1813 only AI worth anything are Tesla's self drive, and Github Co-pilot type shit
CodeRabbit actually has a working product, though!
Rabbit MQ?
“tech ‘startups’ without any product…”
- you can just say “tech startups” imo
Getting highly ranked on leetcode requires techniques that are counterproductive in production. There is a big difference between coding as a sport and coding professionally.
Omg! You nailed it!
Also, just because someone truly is really good at doing something doesn't mean that they'll actually do that thing properly if they don't need to. A fairly classic example is the number of top level speedrunners who have cheated runs, even if they were skilled enough to have conceivably done that run legit.
The Cognition AI team could be the best developers in existence. That still doesn't mean that they actually bothered to make what they said they've made.
Top at leetcode for these bots means nothning lmao. Anyone could google the answers/algorithms required. This bot is just an enhanced search, been that way since the start.
@@hellowill I think he was referring to the founder who apparently has a high rank of CF
I mean, they definitely have potential to become really good. But they need to get actual work experience and do development to become someone useful. And to be a really good software engineer, you need to hone a lot of skills other than raw coding.
Nothing to fear, Devin's just a bunch of humans connected to a server farm Matrix style
OMG, youŕe right, which pill did you take? :D
Sounds like Amazon's AI checkout system
must be the same AIs from Amazon Just Walk out, Anonymous Indians
Wait, you mean like Dollhouse? Like, in the dreaded "The Attic" that's used for dolls who don't comply? Sounds about right. That's why Elon is so big on Neuralink.
Psycho Pass?
Devin is the front end, the backend are a bunch of guys in Bangladesh. Prove me wrong.
"hello devin are you a bunch of guys in Bangladesh" - ask it and wait for an answer!
That explains why the output is so bad
Amazon already innovated on that
you are totaly right
IPUs
A bunch of Indians in a building powering the latest rigs remotely and helping you with all your computational needs.
Until India's populatiom starts to decrease, which it eventually will in some decades. Then we gotta find a better procressing unit.
There is a great video by Internet of Bugs that you should check out where he shows HOW BAD Devin is - he even demonstrates how it's done correctly (with examples).
Yep very good video, ua-cam.com/video/tNmgmwEtoWE/v-deo.htmlsi=HOyvR42XzB89zdvS
We need prime to react to that vid asap
Sadly, it doesn't matter if it's bad or even real... that demo video went viral, and clout is all that matters...
@@reanukeaves I mean, it matters in the sense that all the skilled people can learn the truth and move on, while the dumb investors can lose their money and get stuck holding the bag.
@@4m470I wish.
It's rarely the investors that get left holding the bag or losing their money.
VCs are *REALLY* good at hyping things up enough they can find suckers to sell their shares off to before the truth comes out.
Devin is completely a scam. Saw a video a couple days ago where a guy exposed how the promo video about Devin supposedly solving an upwork task was fraudulent, with Devin making a ton of errors like referencing non-existent files that it then fixed. Also, the timestamps indicate the relatively simple task took Devin about six hours.
It's a scam meant to raise more VC money.
Yeah, you know it's bad when humans could probably solve it in fewer watts :/
The video is from the channel 'Internet of Bugs'
Worth checking out!
All my homies hate Devin.
Gz Up, Hoez (and Devin) Down.
For all my homies named Devin that actually be devin’ 🍻
I never hated something that much.
@@ridass.7137 damn you most have the best government in the world 😂
Devin is based on GPT4 so it'll run into GPT4 problems with hallucination the more context you try and fit into it. The "advancements in long term reasoning" is complete bs. There is no long term reasoning past the context window. The most you can do is store previous text in to a DB with some sentiment about what its about and then reuploading that part of the text into the the context window when Devin thinks it needs it. But this won't happen often because of the nature of how the GPT4 context window is limited in the first place
It is not based on GPT-4 afaik. Its based on a "modified" version of codellama i think
@@Choroalp I’m going off of what I heard. Eventually we will see but it’s very underwhelming right now
@@Choroalpit's based on gpt4 lol
@@crackedoutofmymind-h8m Source please
I don't get this new wave of "AI startups" where it's just some interface on top of chatGPT api.
This might be the "thinest" layer of a "product" I've ever seen.
I'm not saying there's no need for products like this, but how the hell is that a full startup company with investors? It's literarily hackathon level projects.
Welcome to the hype cycle.
What if the video is copium and Devin isn’t a scam?
What are the odds that all these rare gold medalists who work on the project are scammers? Every single one?
@@numalesoybea1348 its bc if you look at the climate, context and evidence, it shows that the Devin team are pulling off a bunch of smoke and mirrors. They're not leading with truth.
On top of that, all the other well funded tech companies like Google and Amazon have gotten caught faking their AI products. You think a few unproven CS guys are suddenly going to change the world building a UI around someone else's product, which is black boxed? C'mon dude. It could happen, but the climate, context and evidence isn't strong.
@@numalesoybea1348 lol what a take
@@numalesoybea1348 very, very high? Like, pretty close to 1.
"Rare gold medalists"... I mean, I dunno how high I'd rank on leetcode, because I'm too busy producing actual work to write leetcode. Same for all the best developers I ever worked with. Ranking high on leetcode proves absolutely nothing about someone being a "rare gold medalist" except that they've got a lot of spare time.
You're just arguing from authority, when nobody had ever heard of these people five minutes ago so they're not even authorities.
99% of the work of a project is the infrastructure around the project. Security, metrics, logging, error handling, outage handing, scale, multi region, deployments, backups, dashboards, support tools, support infrastructure, incident management, documentation. The difference between releasing a product and releasing an enterprise product.
The moment I heard about this I knew what was up.
Crypto Pump and Dump but with A.I.
you are 100% right
finish product is 100% ready sell to customer;
demo is 20% ready;
prototype is 5% ready;
devin show it core function, it replace programmer = 0.1% ready
fund me or programmer die!
AI is the new Dotcom
I'm just glad that soon we won't have to do any more more tedious tasks like coding and we can get back to focusing on what's important, like consooming products and scrolling social media. 👍
This is the future they want for us.
and how are we gonna consoom anything if no one has enough money to afford anything. seems like those in charge corporations are not really smart themselves
If you haven't seen it, I really recommend Internet of Bug's video 'Debunking Devin: "First AI Software Engineer" Upwork lie exposed!'. He points out that Devin fails to complete an upwork task, despite this being part of their marketing pitch.
I mean it's kind of good news in here for actual human programmers in the sense that the snake oil web3 schmucks are likely going to screw up AI coding to the point that the industry takes at least an extra 10 years to develop.
Hopefully it takes long enough for me to retire.
There are strong reasons to say it won't develop. Everyone's so hyped but the reality. Most of what AI is actually doing is mildly improving autosuggest, or giving anime nerds underage porn I mean ah... Wait no it's serious business stuff I swear! My 12TB of "anime sexy 3yo in diapers naked" prompts is totally development work!
@@KevinJDildonik This, unless you're in the business intelligent side of things, computer vision or image generation/ text generation, no matter how much better AI gets, it's always pattern recognition at it's base and won't be anything else.
@@mortvald very, very, very, very, very, very brave of you to say AI will never be anything other than pattern recognition. "Never" is a VERY long time.
Sounds a lot like the apocryphal "no-one will ever need more than 640 KB of RAM".
@@BittermanAndy Nah, this is my field, been my field for 15 years and I can confidently say it. I can recommend anyone who don't know the first thing about AI to look up a video about regression algorithm. It's simple and easy math and does show how prediction models work. once you see it you'll immediately see the flaw baked in.
AI is a tool for specific tasks, been here for a long time and all that happened lately is that big data enabled more complex models but they're essentially the same. AI hallucinates a lot and will always have that 10% to 5% error margin (and this is for world class AI models) it can never be employed in engineering where precision and determinism is everything.
for art sure, a mistake on a draw or bad prompt won't do much harm, for translation sure, a few mistakes won't harm anyone. but for anything that requires precision over prediction? AI will never be the answer
AI was born just like computers were, an attempt to mimic the human mind, now i'm not gonna say we're not gonna pull it off one day, we still have a long way to go and need to understand our mind inner working first but even when we succeed, it'll inherit the flaws of human mind. We're good at pattern recognition sure, but we also fuck up because of it.
It's not a scam, it's entrepreneurship, the name of the game is move the money from the VCs pocket to yours.
How do you define the term "scam" so that Devin isn't a scam? They're trying to sell you something that not only hasn't been built, they don't even know if it can be built.
Mmmhm **thumpthunp** mmmhm **thumpthump** mmmmmmm
So a scam, got it.
It's not stealing, it's entrepreneurship, the name of the game is to move the money from your pocket to mine
@@spicybaguette7706 they ain't stealing anything, the VCs decide to hand them over their money willfully, it's not their fault that VCs pump money into things they don't understand, that's a flaw of the system, not theirs.
"Viagra Capitalist" in chat was criminally underrated.
I am enjoying the next Theranos or the next FTX in the making.
👏
I don't think Devin is going to be that, but I'm 100% there's going to be one.
@@rubendacostaesilva8442 yah Devin is software version of Rabbit R1 😂
Internet of Bugs did a great deep dive on their Upwork demo called "Debunking Devin: "First AI Software Engineer" Upwork lie exposed!"
My 2 cents honestly, the whole Gen AI bubble seems to have been pumped by the wet dreams of corporations having workers fast replaced and splurge into the immense fast returns without having to pay salaries & benefits.
And many of those same workers will be the ones to lose their jobs in the downsizing when this laughable investments soon fizzle out.
I love how it's all crash and burn now, so cool 🎉
To be fair, anyone claiming authentication is easy to get right doesn't actually know what they are talking about. It's easy to get something that seems to work, but much harder to get something that is actually secure. Of course there are services that abstract the problem away but that's basically the point.
I wouldn't trust a junior dev to get it right, let alone one that is prone to hallucinate.
Amen to that
It's not a surprise, OAuth became a thing.
@@3_smh_3and even implementing OAuth correctly is difficult😅. And OAuth is not really created to simplify auth, it's to provide a universal interface for authentication providers
Yeah
"your prompt is scheduled for review"
Aka: shit guys one of the prompts is to make a new website lets get to work.
"Google docs in the streets, Google sheets in the sheets." lol
I started out with rust (and programming in general) when chatgpt came about. The rust compiler seems to spit out rather detailed info about the bugs.
So in some late night sessions, i found myself just copy pasting the compiler output back to the gippity. It was a simple pingpong process for simple solutions that i didnt know how to handle myself at that time.
When things got frustrating, i sometimes thought of a simple program to to the copypasta pingpong for me. Maybe get the gippity to write that for me.
Just watch the chaos unfold instead of actively clicking for it to happen.
Even in those moments late at night with some "why doesnt this f'ing work at all??" mentality in the head, and being too frustrated to do the research myself, i was always sure that this would not lead to the code i wanted to write since the gippity was only a crutch telling me bits of syntax that i needed to get going. Letting it take the wheel even from the very beginning for me, always led to worsening outcomes, deadlocks, version conflicts and unneccessary rewrites
No matter what kind of Legendary Arch Wizard you are, It's hard to take someone selling you future promises seriously.
Funny thing: On their website, at "come work with us" link, you can see they are hiring (human) Software Engineers :D
I know it makes sense, but it's still funny...
if this is in fact the result of VC funding, VC investors seriously need to learn the space. So many of these (maybe all so far?) AI startups are clearly pump and dumps. If I could short these companies, I'd been be shorting every last one of them. They stink to high heaven.
What if Devin is just an AI’s fake influencer and Devin isn’t real in any way, AI fooled us all
Or ai recommended this big scam as a way to get emails when it was asked “what would you do to collect a ton of emails in a short amount of time”
@@elsavelaz Lmfao that would be wild
It probably fooled you and some around you, no? Because any bolivian farmer who never left his small village could spot the marketing effort.
There's also a devin analog that's been under development for longer and is already publicly available called Pythagorea. Also the open source Devika
And MagicHaskeller
"write everything as if it's test first", did I hear TDD mentioned??
I interpreted that as write as if you were doing TDD but skip the tests. However, the implementation should be written to be testable for all parts.
Has prime covered the “debunking Devin” video where the guy shows how Devin didn’t actually do a good job at the upwork task they gave it to him
Whoah! The plus sign trick, I can't believe I was not aware of this! Goddamn, thanks!
WebGPT (the 'GPT' extension by Josh Olin) does better on the benchmark Devin was touting in its announcement.
Devin could be the biggest troll to get executives drooling over just throwing money into their company
My take on this is as follows: If I was a legendary grandmaster CEO, I would not personally write the front end for the company. If my main product was the AI I would dedicate all my best programmers to that task and hire out the job of creating front ends. The problem here though is they claim to have created an AI of the very sort that you would want to hire out such jobs to, and it really does beg the question as to why Devin did not write the front end apps and if he did, why are they so bad.
Devin was coded by Jian Yang after the hot dog recognizer failed.
Influencers are also part of the issue. They hyped it up with no proper research. Cough Fireship
Good luck maintaining whatever crap AI builds. Bitcoin is overly complex and inefficient but atleast it has purpose - druglords can transfer money somewhat anonymously. But this shit is inefficient and has no purpose. You can have best GPT in the world, but if you teach it from all the buggy code which is freely availble it will write even buggier code 😂
I feel sorry for the software maintainers specially open source in the near future, I can only imagine how much crap they would have to deal with.
I love it, hope it keeps learning crap.
Internet of Bugs has a good video going over the specific examples , regardless of their metrics in swe , its the way they are falsely presenting the capabilities .
Devin is already an advanced human. He can scam people
lol
🤣🤣🤣🤣
I dislike the idea of Devin as much as the next guy, but this is an early phase startup building an MVP to do product-market fit. If they are to succeed as a company, it's very likely their product won't be what it is now. Also, the whole "fuck Devin" kickback gets them a bunch of free PR without even having a proper landing page.
Thing is, it's basically impossible to tell the difference between "early phase startup building an MVP to do product-market fit" and "scam", until you actually see the product working. They went public before having a working product. Say no more.
I've been working Voodoo on Devin since the day it was announced. Looks like it's working.
You should definitely watch Internet of Bug's video "Debunking Devin: "First AI Software Engineer" Upwork lie exposed!" which basically exposes Cognition AI for holding back some key information about Devin's real performance, which when examined closely definitely is performing at a way lower threshold than what Cognition AI is making it out to be.
i want to! that is going to happen tomorrow
Legendary Grandmasters? So Devin was made by Bobby Fischer and Kasparov?
nice
Yep, it is. Hype is dominating the media right now. And lots of people that don't understand the limitations of these technologies, get to believe whatever they are told about it. ChatGPT was the catalyst with its impressive, human like communication
Exactly as you said. Once Devin starts writing Devin, everything will change fundamentally.
"Even if I knew that tomorrow the world would go to pieces, I would still plant my apple tree."
Hey! I'm an emacs user from the 90s
What if Devin was the friends we made along the way?
DSA is such a different skill, it's like saying an F1 driver is the best commercial pilot, 'cause did you see how many races they won!!! 🤣😂🤣😂
Somebody's been chugging the kool-aid
The first clue the instant I saw the promo, was that it was a Chinese guy. Racism saves the day again.
Asian American though. They are competitive programmers from the USA.
@@eliana993competitive doesn't mean that they're not crazy scammers. Competent people get away with scamming
@@redheadedmoos1204 I’m not denying that but that person was being purposefully racist so my response was about them being Asian American from the USA as proven by the competition background.
13:09 snake eats the tail situation 100%. not try gas u up but u were on point w that one
also 14:08 LOL imagine being in a conf room together with smone, working on a project or smthn, ... and they start out the blue calling rank and hitting u with that goofy ahh title man i would be SHOOK like am i being pranked?? 🤣
When you typed in your request to fix their site I laughed so hard I hurt my own ears.
I gave Devin a try today. I had the exact same experience.
devin would have deleted the guest and article section in the description
2024, where everything new is actually a scam
There are old scams, aka "fintech" and new scams, aka "ai".
Entire post-modern neo-liberalism is a scam. Ever since they dropped the gold standard we're one scam after another scam in this society.
It’s got a product to help you avoid scams. Are you: a, very interested. B, interested. Or C, not interested
@@Kane0123 that's scam
I made a screenshot on the release day of the openings and they were just hiring software-engineers. If they would have found an AI Engineer they would have hired sales people.
"You get all the code..." Cognition ai needs to study their Terry A. Davis.
6:23 FLIP YOU SKIPPED OVER THE PUNCHLINE
Haha I’ve made a video based on the same posts, good to know that I’m on the same page as the name! :)
A friend of mine had a Professor who said to his class "you better get really good because AI is gonna replace you so fast if you don't" a few years ago (obviously not the exact quote). Devin is the most promising "replacement" and honestly someone with a decent head on their shoulders is gonna be fine. Right now Devin isn't even able to turn a good engineer into a 10x dev, I would concede Devin as a success IF someone who knows the answers to the prompts they are giving him is able to be at least 3-4x more productive because of it, until then id rather have a "go-fer" intern than devin
7:00 the problem is outsiders don't have trained eyes to dicern right or wrong, and many of them blinded by the fear of missing out and greed of being able to do things without learning.
Remember Theranos? Any the medical specialists when asked about Theranos said it was bullshiet, but no investor really pay any attentions.
Yup. While DSA is important and a good skill to have, DSA does not necessarily mean being a good engineer and good at building *systems.* I know guys who can solve hard leet code problems with their eyes closed, but can't figure out that they should version their REST APIs and end up changing end points and breaking clients.
Or in other words. Being good at designing systems and being good at engineering requires being pretty decent to good with DSA. But being good with DSA does not mean someone is good at building systems or being good at engineering.
woah, didn't knew about the plus sign on an email address, that's epic!
This particular "Devin" might be a pile of crap, but "Devins" are inevitable. This is just an iterative LLM using errors as input. Eventually it will be sufficiently fine-tuned to build a viable app from start to finish.
Will take a long time
@@newbieguy2509 fine tuning is even closer to point
Hahahah, Devin powered by 1000 Indian AI or at least 10.000 this time
@14:00 regarding no error handling on a project, there is a 3rd option... they did not get payed enough. Like "Go horse" thinking, just get it to work , barely, then when errors pop up fix them ... Defensive programming is a great skill but eats time & budgets, not very well liked by pushy 'investers' or project managers.
Step 1: Download Devin.
Step 2: Order Devin to make more Devins.
Step 3: Undercut Devin and get rich.
Self driving cars is when we can be afraid. If you get real self driving cars it means they solved the inability to generalize. The fact that self driving cars don't work is what makes me kindof think that LLMs aren't enough.
Self-driving cars are especially hard because you need low latency loop. I think you could already do LLM based implementation for self-driving car if you accept is goes about 1 mile per hour to give it enough time to evaluate everything needed and to come up with suitable option. And even then it might need a supercomputer running a single car.
LLMs have nothing, nothing at all, nothing whatsoever to do with self-driving cars. They're not even slightly related. Not even a little bit.
(Also, you should get a demo of Tesla FSD. It's not perfect, it's not finished, there are improvements that need to be made, there are situations it can't deal with, but if you think it can simply be dismissed as "don't work" then I don't think you've tried it).
@@BittermanAndy I know that LLMs (with typical text based training) have nothing to do with self-driving cars. However, the algorithm itself is generic enough to train from example driving by humans if you give it enough computing resources.
Is it a good algorithm for such a solution? No. But it would be good enough.
Human brain works by having a HUGE biological counterpart to LLM technology and it seems to be enough to allow humans to seem intelligent sometimes.
@@MikkoRantalainen who are these apparently-intelligent human you mention? Don't think I've ever seen any! 🤣🤣🤣🤣🤣🤣
Oh no. I was hoping devin will finally end my way too agile legacy code duct taping every day.
Like I said in so many comments and on stream: Devin is a wizard of oz mvp
This reminds me a little bit of people selling stock trading courses. If people were so good at trading stocks why do they need a course when they could just be making money. Same situation here. If they really had an amazing AI coder they could open a consulting shop and make $10m in revenue per employee
The problem is developers and software engineers are always self doubting because they are so intelligent. Devin might evolve but it will take a lot of effort and great minds.
Usually ThePrimeagen's takes on things are fairly balanced, would have loved to see some other criticisms of this stuff beyond 'a startup made a heavy webapp'. Typically he'd say something like "I could understand that if your focus is on making a tool, you might not rush to make the web frontend nice' etc, but again calling them amateurs or lazy and literally going through the transpiled JS seems out of character for the channel. Looking forward to the next one!
In the words of Jerry Smith “Devin….Devin….Devin….Devin”
It feels like this was a product market fit test, they wanted to see who would care if this really happened
I like Devin. I always felt programming is just menial task to communicate with computer. With Devin, we may finally dive deep into field of development and creation rather than learning to communicate with computer
Current AI just writes code like the average open source/public repo in github. It even generates the same kind of errors that people make. It has its use, but I still haven't seen anything that really looks like a thinking machine beyond superficial use cases.
Viagra For VCs. Now that was an album way ahead of its time.
The second post was written by a door-to-door fortune-teller LMAO
Made about Google docs and Clark? It’s just serverless 😎
Forget whether the leetcode grandmasters can build a working website for their startup. DSA is also very orthogonal to deep learning.
"Devin" is 100% a scam
Grabbing the money is the goal, the only way to build a team and move fast, pre-product. You don't need a polished website for that, and this is just marketing. Empty shell? Maybe, scam, maybe not.
Accelerationist: AI WILL TAKE OVER X FIELD! FOR REAL THIS TIME
Professionals on said field: oh no, anyways
From the exact moment I saw the announcement video I could SEE that it was fake just from the face of the dude describing it
Devin: we built an AI SWE!
SWEs: why didn't they build their own auth?
...
What are you talking about, Atlassian makes great pieces of software like SourceTree and... I don't know, probably some internal tool...
Welcome to Devin's email list, the real project was gathering email addresses.
Main job of AI is to devalue your work only for you to be paid less and work harder. I LOVE SE and i have master degree in it but i am leaving that field and moving to security because i dont want to deal with all that is coming. Gl boys and girls gg and hf.
We can sit and talk about how the current AI tools aren’t great. But the thing is that won’t be the case for long.
Especially not with companies and executives laying off engineers and dumping billions into AI.
Which is a pretty hostile move towards the workers.
Real talk though: It's legit to use google forms for this type of thing... google has a fleet of security engineers that keep that data from getting stolen. Rolling your own implementation (even if it's just collecting emails in a single column single table database), when the best you can muster is the site that they have --- that's a squeel injection waiting to happen, and that could sink the whole company with that one mistake alone.
If there is a breach to google forms, that one company isn't going to be singled out and will be a footnote in a huge list of companies impacted by the breach.
And don't think that developersTM won't be pissed off about being rendered obsolete... hacking attempts of all types are very likely to still happen.
That said, AI for coding is a complete scam, and has been for roughly a year now. The mess it makes and the lack of basic understanding, makes it worst that worthless, as having nothing at all at least doesn't give you garbage code that was trained on trillions of garbage stack overflow answers. The threat of ai, is going to stifle the amount of programmers learning the craft and I'm fine with that... we're full f-off
I have asked to be invited to the Beta. I am fairly well known in my niche, we’ll see what happens with that
So wait…all those folks on Twitter claiming to use Devin were….lyinggggg? ROTF
The virgin 69420 watt GPT vs the chad 20 watt squishy brain
There was a nice tool for python developers. Sourcely, it did traditional stuff parsing your code, using some fancy algos to suggest refactorings, and then correctly applying those refactors. It worked very well for what it purposed to do. An honest tool, a bit expensive, but not too much. It was like a linter, but one abstraction step above, operating on the AST,
For some random reasons I had stopped using it some time ago, and just forgot about it. Then one month ago I decided to give it another go. Oh man! I hope the hell awaits the product folks from this company!
They completely mutilated to product for it to become another "me too!" AI coding assistant. The thing is now completely useless and if you try to use it, create some very short aliases for commands like "git checkout" and "git reset --hard", your fingers and tendons will appreaciate that.
I've been getting scammed ask for something then go and download a model recommended even though it has 180k downloads an then i get spammed with ssh connect attempts.
my setup farts at the 25 attempts per second non stop until i do somethnig, but annoying none the less.