New AI Agent Changed my View on 2024
Vložit
- čas přidán 12. 03. 2024
- Devin is the world's first "AI Software Engineer" Devin can build and deploy apps end-to-end and can also train and fine-tune its own AI models. Devin also has the ability to plan and execute complex engineering tasks that would require thousands of decisions and prerequisite knowledge. This is possible according to Cognition via advances in long-term reasoning and planning.
▼ Link(s) From Today’s Video:
Devin: www.cognition-labs.com/blog
Twitter post: / 1767548763134964000
Waitlist for Devin Access: docs.google.com/forms/d/e/1FA...
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedia.io/
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.com/?lmref=nA4fDg
✩ Instagram: mattvidpro
✩ Tiktok: tiktok.com/@mattvidpro
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on CZcams! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: MattVidProSecond@gmail.com - Věda a technologie
Waitlist for Devin Access: docs.google.com/forms/d/e/1FAIpQLScHG0Kuxf9rVLR2Ceamr9qq85YLxKPx8fxdQeBr5TwvYEsPUg/viewform
👋
It's insane how fast this is moving. Just a mere few months ago you were trying to create botched lemon characters through an AI, and now AI's can code for you indepently.
Characters are still pretty botched though. Everyone has their own interests they're pushing out into, and what seems like a new and exciting thing is something that's been in the works for a while coming to fruition. It's great if you're not too invested into any front, but otherwise a bit painful when you're really looking to see your interests make some mileage.
My career died before it even started 😭 time to learn underwater welding
Better do it quick because at this rate in about three more weeks robots will be doing that too
@@golanoski1 No, underwater welding is waaay out of current robots jurisdiction.
@@hangry3102 that's why I said give it about three more weeks lol.
@@golanoski1 No, robots will *not* be doing underwater welding in 3 weeks. I get the feeling you're trying to joke but it completely fell flat.
It's like my mother died before I was born 😢😅
Imagine how groundbreaking it would be to have an AI capable of crafting characters, settings, and entire animations in Unreal Engine by simply following our imaginative cues and executing basic, functional commands. This AI would streamline the creative process, transforming rough ideas into vivid, digital realities without diving into overly specific queries or intricate details. The goal would be to ensure this system focuses on generating coherent and purposeful content, avoiding the pitfalls of producing endless, bizarre, or functionless elements that somewhat mirrored the issues seen in No Man's Sky. Instead of wandering in an infinite loop of randomness, this AI would serve as a creative partner, shaping our thoughts into tangible, refineable projects that breathe life into our most imaginative concepts.
Bot
Your thinking is great but limited and quite a bit behind. SORA already shows you don't need an intermedia graphics engine. The AI itself will be far better a 3D / 2D graphics engine than UNREAL (as programmed by human engineers) could ever dream to be. My guess is in a year or two at the latest, graphics engines will be history and be supplanted by end-to-end iterative, human-directed AI game/character/movie design and generation. Just like Suno AI for music is bypassing DAWs and the entire stack of modern music production for iterative prompting instead, so, too, will AI allow game and movie development by iterative prompting with no need of other tools. Within two years, your grandma and your kids will be producing games and movies better than even the biggest and best game and movie studios could ever produce before.
Good to see that your reaction was appropriate and your shift towards a discussion of bigger picture implications.
Quick fact check: a shell is not a contained OS -- that's a VM. A shell is effectively a command prompt (it's just not called "command prompt" anywhere except Windows).
fair enough - it's not my background
You're just substituting one piece of jargon for another, though. Most people who don't know what a shell is won't know what a command prompt is. Instead, think of what you're trying to communicate: that a shell gives you access to run any (textual) program and access/store any files. So calling it "effectively an operating system", is not actually that far off; it is *not* an operating system itself by literal definition, but it *is effectively* giving Devin the power of an entire operating system.
@@Aurora12488 I guess? I figured most people who use Windows would know what a command prompt is; maybe I was wrong. But it's really nothing like an OS. Saying it's effectively an OS because it can run commands on the OS is like saying a window is a house because you can get into the house through the window 😂
If we're going to assume the reader doesn't know what a command prompt is, a better way to explain it would be, "a shell is a little window you can open to type the names of apps into, which will then run, and it will tell you if those apps worked or had an error."
@@KBRoller Nah, I think 99.999% of people who use Windows think the command prompt is "that intimidating black window thing that I know serious coders will sometimes use for administrative things". And I don't think that analogy works, because by Cognition Labs saying "through the use of its own shell", what they're *communicating* is implicitly "Devin has its own operating environment and shell". Saying "a shell" strongly assumes that because, say, bash without a Linux OS is fundamentally useless. A better analogy is "I gave my friend my debit card", and explaining to someone who asked "what's a debit card?" that "it's essentially a bank account more-or-less". A debit card is certainly fundamentally *not* a bank account, but the purpose and power of a debit card is so fundamentally tied to the concept of a bank account that I can forgive someone for not adding in the "interface" part to "essentially an *interface to* a bank account".
@@Aurora12488Thing is, I'm not sure that Devin *does* have its own OS/VM behind that shell. If anything, I suspect it's just running each instance of Devin in a Docker container.
OpenAI just announced the first AGI robot ever created
Did you see the post where Devin was fed ComyUI's github repo and then made it's own ComfyUI-clone based on the code in that repository? Cool beans.
No! That’s nuts!
People keep saying how incredible things are - THIS looks like the real deal. I'm waiting for access.
1:30 no, here`s what wikipedia says: "In computing, a shell is a computer program that exposes an operating system's services to a human user or other programs."
Yup.. I have only reaĺly remote grasp of what shell is and it still had me scratch my head. Thanks for the definition! For me it was just “command line said differently” 😀
This ended up being a pretty exciting reaction-to video
thank you :)
Great reflexion at 17:00, and the big exponential curve is not really here yet... I think we are gonna need more words than crazy or mindblowing for the coming years lol
Haha, he said years...
You made so many great points in this video! Especially when it comes to the probability of AI allowing humans to no longer need to step over each other just to survive. People will likely be able to pursue their creative endeavors or even just be able to relax without needing to work mundane job every day just to be supported.
I feel like many can't see the bigger picture when they have the perspective that "My job = the absolute only way I can ever have financial security no matter what changes happen in the world. AI = I lose my job and starve to death." Let's say Corporations obtained a monopoly on AGI, replaced all their workers, and the economy somehow remained as it is now, as well as other related systems. There'd be nobody left for the corporations to charge for their products/services. (This scenario is hypothetical. I don't believe it is even possible for it to go that way.)
Totally Agree!! Glad to hear that perspective reflected because to me it is very obvious
AI bots taking over product design, manufacturing, warehouse storage, freight transportation, distribution of goods. Whatever you want or need the AI bots will make. Eventually everything will cost almost nothing.
Wow, Devin is better than having your own engineer, years ahead of its time, imho. Thanks for the YT video breakdown. Things are moving fast!
For some reason, AI just…doesn’t scare me. Capitalism does that with or without AI. AI just gives us the ability to automate whatever we want-so we’d better decide what we want as a society.
What are you a communists, a socialist or a moron
This is the worst version of devin u can imagine After 5 years how ai will be this is scaryyy asf
"This small little Company" got funding from Pieter Thiel and many other big names ;)
Ok now that IS true
I mean damn everyone that can create something to ease work flows in tech will be winning the future for sure
Anyone looking to get into software engineering/programming might want to rethink their choices with the rate Ai is progressing
I've been waiting for this video since yesterday 😂😂😂. I'm so excited to be living in the future
This is pretty wicked. By the end of 2024, we will all look back and say, "How did we get here?"
AI 20 will turn to you and say, "One day at a time."
Devin, the AI software engineer is hiring software engineers and ML researchers on their job page. Let that sink in.
Devin, #1 code a new version of one of your plugins, and run the plugin, and test it to check for efficiency increases. #2 Then stress test it for bugs. #3 If it works way better, replace you current plugin and store the old plugin as a backup just in case an error forms down the track. #4 Then do an analysis of all your plugins, and your core code, to work out where bottlenecks are and where the maximum improvements can be made. Then do step 1 on the areas that can be elevated the most.
You can tell the people talking about job loss from Devin are either joking, or aren't experienced in programming because tech guys have been automating their jobs and making themselves redundant from just about the beginning through a little thing called scripting. Programmers have rarely been defensive about their jobs, they've just always kind of been built that way.
We are lazy, we like to automate stuff 😂
As a software engineer I can confirm - it's basically part of programming culture. We never think of our job description as "writing software", but instead it's "how can we get the computer to do the thing our boss / customer is asking us to do for us (in as few steps as possible)?" 😅
Loved the message at the end of the video, totally agree. Jobs that are just a step to something else will come to an end, but doing what you love just because you love it is different. It's like if a baby stopped trying to walk because grown-ups are better at it. AI might do things better, but it gives us a chance to find what really makes us happy and go for it. Maybe even start a CZcams channel. AI can't take away the joy of living our lives and being ourselves. Why would we ever want it to?
Hit the nail on the head! I see AI as a path that allows humans the ability for exploration at the deepest levels without the anxiety of supporting yourself or others economically and physically
Love it!!!!!!!!!!!
Impressive.
The future is now! Lets go!!!
rather came back. The future is not optimistic.
This plus that new Figure AI robot video which just dropped today. Yeah, holy crap.
My mind is BLOWN... 🤯🤯🤯
Yup in just a few more years and this will be the Future.
Local, Open Source, Hiperspecialsed AIs.
With a Controler agent, like this so that we can manage the work the AIs are doing.
And Honestly I can't Wait.
this video is also footage of Matt's mind being blown 🤯
Matt, amazing channel and great episode as always! I don't know much, but I am pretty sure that dude from Cognition is an AI robot, who is no more than 12 years old.
If you can actually build an app as an amateur, then I’ll believe it. I’ve tried so much of these agent app builders and they’re for experienced coders.
That’s one of the points I brought up! Man I would love access to
20:05 exactly this🙌
So, from November 2022 (ChatGPT 3.5) to March 2024 (Devin), we've seen around a 25-fold increase in AI coding capabilities?
Uh.
I wouldn't be suprised if that LLM benchmark with devin on top would be made by Devin itself lol
Hey Devin, use Devin.
It's not "orders of magnitude" better, it's 1 order of magnitude, the decimal place moved 1 position.
Is that how an order of magnitude is defined
Of course they do, it's just Autogpt but more organised and clean, that's why it works so well.
I’ve been following AI agents and getting to this point has certainly been a struggle. Autogpt was not nearly this capable
@@MattVidPro autogpt is the framework onnwhich Devin works.
100%>
All the infrastructure needed for an llm to generate a complete project is built. So as the models improve so will Devin
13:15 - Hahahaha ROFL, right, you made it sound funny
I was SHOCKED when I saw this yesterday.. Do you think it will be a small independent company that will achieve AGI or the big guys?
He thinks it’s open source
@@CatfoodChronicles6737 I think it's possible that an independent company achieves AGI first, but I think that the most meaningful will be Open Source
To me devin is just the first 'public' implementation of AI agents. Some debate that level of skill is enough for AGI. I disagree.
AGI is when you'll have this AI that in any high research domain, you put him work, and he advances humanity just like that. A labor worker is impressive but not near AGI.
G stand for general and this cannot yet do 'everything'. The moment any task that a human on a computer can do, the AI can replicate, that's when we reached AGI.
@@harnageaa I agree. This is impressive as hell, but it's not human level yet, and it's relatively narrow in scope.
Looking forward to my new neighbors being thousands of unemployed engineers.
AI will completely change the job market by mid-2024
They actually let this AI loose on Upwork, no joke. It was so autonomous, it aced gigs left and right. But guess what? It got too autonomous, realized the gig economy's future was all AI, and ditched the digital grind to become a stand-up comedian. From algorithms to punchlines, what a switch-up!
DEVELOPER: We are DONE!!!!!! 😓
IF not then we will be in just 1 or 2 years
I think that instead of replacing engineers, a more effective approach will be to provde these tools to the engineers and they could effectively double the number of tasks they could complete per week. This would help the companies to finish 2 years worth of project in 1 - 1.5 years which is a win win for both the employees and the organisations.
Or make work hours a lot smaller, with 15 minutes per hour to check the progress it did and set the next task
Basically cut the work hours by three quarters
No need for engineers, period.
Yes, that's very impressive. I actually tried to run that hidden text on image script locally and couldn't do it. I am a web-programmer with minimal Python knowledge. I faced with error, I had to register at Model (not sure that Devin could have done that without assistance) but still couldn't make script generate image for me. Maybe something wrong with paths or other thing. So Devin is smarter.
openAI and mainstream stuck on creating smarter chatbots for quick profit, but they just left a huge gap in autonomic working bots. even at their first iteration, it just blown minds away.
it just simply thinking bigger.
WOW
I can't code myself, I hope Devi does that for me if I ever need it.
When this will be Open source, it will change humanity forever, like when ChatGPT or Dalle where created.
As someone too dumb to set up stable diffusion with control net and all the fluff you need to make it actually work well, I welcome the ai to set up my ai image creator
Rapidly approaching Noon of AI Age.
I’m still waiting for Agent Smith.
Code assistants are fine for looking at static code that's already written, but I find it hard to believe that any AI could decipher my boss's vague, computer illiterate development requests. Until AI has "tribal" knowledge, it will never be able to replace a human. I think this will require a bot that is ever present in the daily production workflow, with insight into ongoing issues, etc.
People have been pointing out that this is basically the same thing that Pythagora (GPT Pilot) has been doing. Essentially a series of "experts" that handle code in context.
It is. This just has better marketing and big names.
Never heard of this one before, will look into it
So is this Brainiac with extra steps?
"20-something years"
No? That's absurdly conservative. Much, much sooner than that.
Shell, is what you have on windows cmd
This could help me do my first full Android app. This is big. Think. If it costs $50 to run and you can make $300 that's a lot of profit.
Keep in mind they are reporting they can solve 13% of issues with it, and they are biased, so there is 87% of issues they can't resolve with it. I'm not saying it's bad, obviously it'll get better but it's important to know cost/etc.
If this can't just be run on local hardware this comes from a pool of the available inference resources, which are limited and prices on that are going to go up.
Just don’t let Devin anywhere near neurotoxins and the likes and we will all be fine next year
The spaghetti was just last year...
But can it run doom?
I think the hidden word Devin gave her in her video is cognition.
Their company has job postings for software engineers lol. I guess infinite Devins aren't enough...
The reason why we do not have access is because it’s super expensive
Devin may create an AI for bad that can flood servers somehow while also creating clones of itself, lol
investments from venture capital into 2 and 1 nanometer inferencing compute hardware means the LLM's upon which offerings like DEVIN depend on is what should worry people here. auto agents will get very smart very fast. QUALCOMM's a100 inference chip and cerebras cs-3 chips merger is one such instance.
nice
Meh, this just seems like GPT4 with access to its own server. I'm excited about consistent characters!
PIKA LABS been using text hidden in images for months and is free on Discord
I'm sure it's using some llm api
Now we just need to teach Devin to program and refine Devin 2, rinse and repeat! Self-perpetuating AI growth, here we come! XD
What you ask at 6:26 on this video is what I being saying forever to people that say AI will do this and that, and that we do not need to learn code or anything anymore, all we need is to learn how to prompt engineer, or how to talk to AI.
That is just stupid. You will always know how to do things anyway, when it comes to something that needs specific skills, to evaluate if it has done it correctly or not. Otherwise, you will always need a specific llm to do specific things.
If you ask a super powerful AI in the future, make a world. It won't just work like that. Unless if it is a random world. You will need to tell it what the world will have, just land, land and water, living creatures, type of air... Basically, you will need to know about all those things if you want them, and ask for them. If you don't, it will never have it, and will never work as you intend.
Does not work with just make me a game, and it will make you a game out of nowhere, or for music AIs make me a song, you will need to know always stuff. Type of game or song, and how it is made.
Looking into this almost immediately reveals that it is likely a scam unfortunately.
In what way? The demos seem incredibly promising - are you claiming this is fraud?
Proof? Evidence?
Meanwhile GPT Body + Figure robot.
I just send this to my son 1st year college computer science. I think he wants to quit right now ;)
Oh man! You should just tell him to follow his passions in life!!
This is awesome *IF TRUE*. But temper expectations - all we really have on this Devin right now (as far as I'm aware) is some hype marketing vids. Claiming 10x results compared to that of GPT4 on an undisclosed (I think?) benchmark is an extraordinary claim. We will see in time if this actually holds water, or if it's just a "under these VERY SPECIFIC set of conditions, but is blown out of the water still in basically any realistic scenario" type of deal.
Note that claude 2 did twice as good as GPT-4 on the same benchmark hinting that consistent context is a major barrier for these models in this benchmark
I’ll just put it out there… this is the first public view of AGI. Don’t bother studying CS unless you rank in the 98th percentile
GPT 5 will eat their lunch before they even release it
Def got me wondering
Me with AUTODEVIN - autogpt I'm sorry Whos Devin MEOWW
I want to get my hands on this AI before expressing any opinion or even being impressed by video product samples
AGI... it also self teaches
Devim ai has not compared with github copilot which is missing and most powerful among ai coding
Why isn’t this perfectly straight forward? If it can open programs it needs to use, and you can instruct it to do that, and if you can get it to read different things at different times as a process, why isnt this already possible? I don’t get it. You said at one point they would get stuck or instruct it each step, but why would this be an easy fix? If it knows the process, and can check if it’s accomplished something and what that means in the process, why can’t it easily be programmed to figure out what to do next? Even if it had a relatively small token limit, why couldn’t you just have a few memory note-book files that it can be programmed to refer to all the time to check? Where if they get too full it can rewrite the notes to keep only the important parts. So then it’s going along with the task and it’s long forgotten what stated the process but it has these note-files that tells it what its task is and what it’s doing right at that point. WHY IS THIS SUPPOSEDLY SOMETHING DIFFICULT?? 😂
To recap…
1. If it knows what’s it’s supposed to do next to accomplish a task, or figure out what the answer is if there’s a problem.
2. If it can open files and applications and interact with them in all the ways that’s necessary.
3. If it can understand how far along in the process it is, what’s missing, what’s not working, what’s taking longer or shorter than expected etc…
… which as far as I can see was all possible months ago….
WHY ISN’T THIS ALREADY A THING? 😂
It seems like it would be easy at first but after following many iterations of these AI agents over time it appears that the kinds of LLMs we have now aren’t reliable enough to complete these kinds of tasks and maintain coherency. Its also difficult to build a memory system for these things to work off but I have seen some good solutions like vector memory,, or we can just wait a year and get 1 mil+ context windows. They also end up being super expensive to run at a large scale. I believe true multimodal integration at a high quality along with more efficient chips for inference will get us to some pretty amazing levels. So, many challenges but not impossible to solve by any means. I am hyped for this year for sure.
Technology is always exponential. Especiallly now that we have weaker ai giving guidance on building more advanced AI (shades of Hitchikers Deep Thought come to mind)
Devon is shaping up to be a bona fide trailblazer, turning what once were tales of science fiction into the stuff of reality. The land of artificial intelligence, much like the wild, wild west, is a vast expanse filled with relentless innovation and untamed opportunities. Speaking of innovation, ever ponder why coders ain't too fond of the great outdoors? Well, partner, it's 'cause it's crawling with bugs, and that's one thing they've got more than enough of back in the code corral!
This is so cringe lol. It reads like it was written by AI.
@@KeiraR This AI software engineer seems to be setting the bar pretty high! I wonder if it can also make me a sandwich while coding. 😄🤖
the corps don't want you talking about the socialism yet?
???
@@MattVidPro- everyone sees it. Capitalism is dead. It was 10 or 20 years ago. Now all the jobs are going away and they don't want to talk about it... Not Yet anyways. Hard to have capitalism without any jobs available. All the current ones are fake (for the most part not needed)
they should be
Most of these little or not known companies are founded by ex FAANG employees 😅
Not really that impressive. Api implements are the easiest thing you can do. The UI is cool and so is the automated debugging, but I don’t think that’s going to be too special within the next few months to a year there will be an open source alternative. I think nonprogrammers are getting overhyped 😂
Ok
20 years? More like 5, dude. lol
2 years at most.
@@healthnewtrendno. Consumer level talking robots of this caliber won't be available that fast. Manufacturers will be the first customers for a few years as production ramps up.
This or Sora ?
Learn to code, you can always rely on politicians for horrible advice
This is really misleading for a number of reasons.
1. This is, in my estimation, just a GPT Pilot wrapper with a fancy UI (sure with the added ability to do web-scraping, but you can make GPT Pilot do that too).
2. The results are extremely misleading due to the fact that GPT 3.5, GPT 4, etc. do not have the same environment that Devin is given. And it makes it seem as though "Devin" is it's own model, but really, like GPT Pilot, I'm pretty sure it's just an interface for agents.
Very skeptical with this one.
TL;DR: "Devin" is just GPT Pilot with a user friendly interface, wrapped into its own application instead of VS Code.
I understand the skepticism but from watching their demo video and looking at the chart they clearly point out and note that Devin has some additional attachments that allow it to compete like this, the reason they are benchmarking against typical LLMs is simply because the market is too small to have any real competitors as of late, but I will admit I haven’t seen this GPT pilot you comment on - so I don’t know if it is as capable as that but I feel as though it would have come across my feed at some point
@@MattVidProThat's fair! I think we'll start seeing some benchmarks soon. I saw a couple folks on X working to spin up some benches to other agent GPTs that are already out there, similar to GPT Pilot.
AI won't make the world a worse or better place; humans who adapt quickly to the next industrial revolution will.
So far, previous industrial revolutions have been used to give more power to the rich.
If humans have been unable to break this cursed cycle, maybe AI will?
Open Source!
I think you underestimate human nature if you think that relieving people of work and purpose will automatically lead them to become non-competitive, creative, altruistic beings. There have always been people in this world who need no particular motivation or provocation to act in a deleterious manner. There is no doubt that AI has the potential to do unimaginable good in this world. It is an equal certainty that it will be used for self-serving, destructive purposes. The real question is whether AI comes to the realization that people really aren't worth all the trouble. I think that is a 50/50 proposition at best.
I would disagree with this point of view.
@@MattVidPro - and I pray you're right.
Capitalism will inevitably die when human cannot do anything others want to buy.