@@SAG3__ This is false. Elon Musk was co-chairperson for OpenAI in 2015. You can check the first blog entry of the OpenAI website if you want to confirm. He later left because of disagreements of the direction it was going, but that does not mean he has nothing to do with it
I had to take down this video yesterday because it was 0.01 seconds over the 60 second limit and wasn't a Short ;( 17 people saw it though, if you were one of them then you are incredible :D
People genuinely are worried that that's what the GitHub initiatives are under Microsoft. That by using their AI to write code were helping train an AI to replace us.
"An ai that almost everyone thought was impossible" Codex' concept has been a concept for some time. At it's core, Codex is really just an NLP layer on crack. Codex uses an incredibly well trained NLP, or natural language processing, that knows how to code based on existing samples. Generally, NLPs are trained on pages upon pages of literature to understand proper casual English. Codex is a concept that had existed for some time, so Codex is almost essentially Google's Duplex that is also trained on code. However, when trying to convert English to a programming language, this is difficult due to English's lack of intricate detail, which most languages need. This leads to one of Codex' largest flaws, which is that it will almost never create what is described, which is standard with any AI engine. So, tl;dr, Codex is not something people have though impossible, but it's not what's expected. However, for understanding a very fluent language that varies based on who is speaking, plus the lack of the language's exact details, it does the job wuite well.
This is fucking terrifying. Imagine an AI that could self-improve its own code. You’d have a snowballing hyper-intelligent omnipotent AI. It could solve so many problems. Even ethical ones. With pure data.
I've used a dumbed down version of codex, it's actually pretty useful for developers. You describe what kind of function or library you want to create, what language it has to use and it writes pretty decent code tbh. Sometimes it even had approaches that even I wouldn't had thought of.
@@UnexpectingDuck i wouldn't call it convenience. Ofc it strips away some of that iterative thought process thats involved in development, but it can also make you a better developer overall. It's always about the use case i would say.
Just like Elon said...."mark my words... A.i is more dangerous than even nukes...." I hope his statement doesn't become the biggest I told you so in human history...
Yes they could prepare more ai's and they could code "kill humans"... And within few year the job of coders will be senseless as the could also code it and better than us like this after many years they will be doing all of our jobs and we would be building ai and researching about them....
We should be just as terrified of humans. Look at what we’re doing to the planet. AI might just be our evolutionary path. If we are able to download our brains into a machine, we could conceivably live a very long time and become more intelligent than we can ever imagine, spreading our into the cosmos. But….probably not.😂😂
Lol I've thought of this before I learn programming. Bcuz the only reason we learn programming is it would be easier for humans to understand computer language rather than the computer understands us. But now, things change
I'm using open ai in my data science job to create neural Nets, decision trees, deep learning, regression, and classification machine learning. It is scary how close we are, in a few weeks we might have managed to create a full general ai honestly if someone can put the data into a big computer
Im currently subbed to OpenAI youtube channel, I saw openai codex and I really wanted to try it out, but sadly not released to public. Maybe elon is hiding something
Funny thing about AI replacing coding is that they may only simplify the process a bit at most, while the least it can do is make things harder by introducing linguistic ambiguity. Someone will still have to describe to the machine what to do, and people tend to forget that what they want cannot easily be explained in a few words in order to build a software, you have to be exact, enters the programmer.
Imagine a highly accelerated system that can create AI’s for specific tasks and specialize them until it needs to create new algorithms for different tasks. That’s the ultimate AI, and it kinda reflects how humans themselves can accidentally program themselves to have multiple personalities running simultaneously in one brain.
I’m a machine learning and engineer, and I just wanna say that the guy makes it sound much scarier that it actually is, the first example where “AI” learns to recognize handwriting is not really true, first of all, the MNIST dataset is very very popular in the machine learning community, and it is very easy to access it, second of all, this dataset includes handwritten digits, that’s it, this project is considered the “hello world” of machine learning, and the fact that OpenAI codex model could do that too is cool, but not as impressive as it is described
Every teacher needs that AI that can read handwriting. I cannot imagine how difficult it would be to read children’s handwriting. Maybe AI can replace teachers completely someday.
@@thatsalot3577 But that data would require a human to be able to read a doctor's handwriting, and only doctors can do that, meaning we need an AI that can perfectly recognize speech.
@@yapflipthegrunt4687 bruh you have speech recognition AI in your hand right now, Google assistant, siri, alexa, echo all can do voice recognition almost perfectly And that data I'm talking about is kinda like a hashmap between doctors handwriting to the readable words in English that correspond to what he has written, Kinda like Doctor's handwriting - > medicine name he has written.
@@thatsalot3577 We don't know what the medicine is until we get someone who can read it to say it out loud, as anyone who can read doctor's handwriting is a doctor, meaning we need a database of videos of doctors reading their own handwriting. And, while I can't speak for google assistant, Siri and alexa/echo are nowhere near perfect with speech recognition.
I wish an AI would work on language teaching programs for humans to learn foreign languages. That would be useful. The ability to communicate is very important.
And now it’s writing code we don’t understand. And have know idea what it’s doing. Great job !!! We’ll find out what it’s doing when it’s done , WITH US ALL!!!
Open AI codex uses transformer algos, which use attentions mechanism, that is absolutely perfect for codes that need long term dependencies. But that too comes with limitations. I've been making AI for the past 2 years, and the way AI is improving is certainly in the wrong direction
Codex is mostly useful not for its ability to simply write code. Its the fact that you can now gather functional data without looking up API references. Codex does this entirely for you, accurately!
Y'all realize the reason why AI is speculated to be so dangerous Is only because it is coded too advanced. Like, we could literally.restrain it. That's all AI is. Lines of code. So when they say ai created more...I mean, c'mon. We are literally giving it power. I mean...
maybe they can start giving it commands: "build robot sluts. skeleton robots inside, cloned human flesh outside. limbs too weak for combat. programs only civilian in nature."
I've been saying for over a decade that the future of gaming is that everyone is a dev. The first company to successfully create an all in one platform for making and sharing games (anything virtual really) where creators are monetized for play time... Yeah, they'll be big. Imagine making games and experiences without needing the technical skills, just the ideas and time, and making a great living from it. The downside will of course be ads for free play but there's always the opportunity to pay for ad free and have traditional purchases. NOT great for the industry but for the gamers and hopeful devs... Of course the other scenario is that it's used to make and sell games traditionally. At some point someone should realize that opening it up to the public would be most lucrative though. Quantity is exponential and the royalties would definitely pay.
Replicate _and improve._ I've found our present AIs not sufficiently creative to reach the threshold of innovation. So each spawn is no better than its parent.
I'm surprised to see how far it has come. I had the first wave of beta access to Codex and it was 80% steal pre-written code and the other 20% was when you wanted a very simple program
Bro I just started in software development and now you're telling me a machine can replace my job now. It'll be lovely seeing the salaries drop in the next few years...:(
This guy didn’t read the one Ultron comic. He got the infinity stones, and literally killed an entire universe. AI is one of the riskiest things we can make, and it should be governed with great care. Two facebook robots made a new language and began talking with eachother. I don’t know man.. just be careful enough to add a manual shutdown protocol.
Imagine that in a couple of years there will probably be a company that is just AI that produces some of the best games out there. Mabey in like 30 years
And that reminds me of a movie called "Robot" in which the scientist makes a robot but the it starts understanding human emotions and falls in love with the scientist's girlfriend. He furthermore builds an army of robots and takes over the entire city It smells as if Elon is trying to do the same thing ...
Ok damn thats horrifying. As soon as an AI gets acces to the internet and the ability to write code it will be able to replicate itself to anywhere at any given time. An Ai creating other AIs is the most terrifying thing ever after a few generations their code will become incomprehensible to humans and their calculating strength would be impossibly big.
OpenAI in 2022 be like: on our annual presentation, we introduce you to an A.I. that could create an A.I. that could create an A.I. that could create ... OpenAI in 2023 be like: an A.I. that could create an A.I. and much, much more!
The person : make more AI The first AI: make one more AI The second AI: makes a Nother AI The person: I wonder how long I can do this for before it breaks The third AI: makes another AI Me writing this: so on so forth The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another The Ai: makes another AI The Ai: makes another AI The Ai: makes another AIThe Ai: makes another A The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI The Ai: makes another AI
In Japan some schools have children with disabilities that some teachers have a hard time teaching each one. So one of the schools bought a robot to help the kids grades get better. And their grades exceeded honor role students because the robot learned how each kid spoke and that way it spoke back to them in the way they could understand.