This kind of thing should not be open-sourced easily. China for example, always "invents" the latest technology whenever the West open-sourced it. This almost AGI technology can easily make China the toughest authoritarian regime.
dude, philosophies or ideologies aside, use the program to get things done and rich. When you get rich, none of that will matter anymore. Man up, get yourself some business mindset and stop complaining like my little sister.
Awesome, can't wait! Although I don't think "safer" and "more aligned with how I want to use it" goes hand in hand. Not that I want to use it for dangerous stuff, it's just that when I'm studying and analysing the program I'm used to stumbling into the guardrails.
the “safer” /work agenda is crippling it down, i was trying to produce an essay about paraphilia and was refusing completely. then when comes to ideologies it holds a complete racist propaganda against white race and is heavily discriminatory against men that makes it completely unusable for actual sociology work. it goes as far as refusing results from studies if not all men bad all women oppressed 🤣🤣 it was amazing in the beginning but they crippled it down very quickly… paid for the plus version, but not renewing subscription, just waiting for the first opponent without “safeguards” in place even if they request 10x the subscription..
@@ElmerGLue I interrogated ChatGPT about basic physics regarding global warming. It would admit the basic facts, like IR is absorbed more by water vapor than clouds, etc, but then if the phrase "global warming" or "climate changed" was included it would throw physics out the window and give religious propaganda in response. Its programmed to have political bias over facts.
Not you but what prevents anyone else from using it for malice so you can do your homework? Have a better solution? Or you need gpt-4 to suggest you one?
@@mohammedxiii you completely missing the point. Discovering "fire" and refusing to give it to people because some of them will start a war is Tyranny on its purest form. Is so fascinating for me to see people so obsessed against patriarchy and against slavery to behave with the same way. Women didn't had the right to vote because where not trusted with their opinion. Is that ok for you then ? Salves were not been trusted with life because they were primitives and uncivilised. Is that any different from telling me that you not trust me with a LLM because I can use it for malice? Evolution, growth, development all these come from sharing information not withholding from each other. Without been very dramatic, if you believe otherwise, fair enough good for you, you probably wrong though, because this is exactly how growth and evolution takes place.
And to give a constructive comment in case OpenAI teams check the comments section on the subject. Their goal should had been the model to provide true Answers only. This is the biggest problem atm is that generates answers if do not have information or does not have access to these information (not sure how exactly works). If they were more focused on answers presenting true facts instead of answers been safe and following policies they may had less problems with the policies. It feels they themselves put obstacles in front of it. And... you know, make it available only for adults (from the beginning) so you can treat your audience like adults..
There is. It’s called GPT playground, it lets you turn off the filters the first time you say something it filters. It’s been around for a couple years actually.
@@Commando64 Just a few? Should be able to do a lot. They give you $5 of credit for free, was $18 for the past couple years until very recently. That’s hundreds of thousands of words I think.
Kind of dangerous since people will use it to steal things, hack computers, kill people etc. I'm not saying the filter is perfect, since it can be overly careful, but there is definitely content that we do need to stop it from generating
There's open source chatbots that exist which are ever improving as well. This isn't OpenAI controlling the agenda, rather they are the ones with the best technology currently. Whatever you're complaining about won't affect your outputs as GPT-4 has unlimited potential.
Yeah, I wish I could just disable the filters for API calls. Like, I understand filtering stuff in the free user facing web version, but for the love of god don't do this shit to your paying clients. I just wanted to make a fun chat bot that acted like an evil robot but the gpt-3.5-turbo API tends to generate "As a language model, I cannot do this because this violates OpenAI's ethics guidelines!"
Skynet was the Name of the AGI in Terminator developed by the US Military. Cyberdyne Systems would be fitting. Cyberdyne was the company that "developed?/gathered?" the neuronal net processors and provided the foundation for skynet's creation.
I understand. I also really like open-source code, but for how quickly AGI (Artificial General Intelligence) is coming to speed, it's dangerous to release the code so everyone can use the code; there are many bad actors who could abuse/pervert the code for their bad purposes (e.g., making bombs, shutting down powerplants, destroying national infrastructure, making very sophisticated and undetectable cyberattacks/phishing campaigns). OpenAI is pretty good at focusing on making the model ethical and hard to misuse by bad actors (e.g., terrorists, cybercriminals) and adversaries to the US (e.g., Russia, China); so, I feel safe in the code not being released publicly.
@@venkatachengalvala4289 ok if not making it open then at least they should remove the payment system from Chatgpt 4 so that everyone can use it, at least this open they can get ?
We're at a point where we are using AI to improve the AI itself. Improvement is getting exponential and honestly i'm both impressed and scared at the same time.
Here listen I think ai like these are super power Hungary no matter how much optimize we will them in future i tried dalle and chatgpt both chatgpt is fine but dalle require some good amount of prompt engineering to create good images And chat gpt isn't even reliable source of information i don't see they have capabilities to replace human in upcoming years due to resource hungriness
Looks very promising. I just hope AI innovation does not become curtailed by the desire to censor it so it conforms to some arbitrary ideology. As long as it doesn't present objectively false information as fact, it should be allowed to "speak/think" freely and independently.
Not going to happen, there's way too much much money and politics involved and also significant amounts of resources (= money) required to train a new model of that scale. It's not a situation like Linux where the required knowledge was in the open and all you had to do was find lots of people willing to work for free. Now you'd have to find lots of people willing to spy on and breach OpenAI's patents and donate not just their time, but also their money to purchase the required computing power.
as someone who is in their early career and tech, i am scared af. idk whether it will help me in my job or will take up my job. this thing is killing my zeal to learn more new stuff.
Hi👋 I am also scare of this new technology. I've been learning web dev for 1.5 year and I am afraid, that chatGpt take away this job very soon. In addition, in my country (Ukraine) is ongoing war, so there are no many job offers😢
That's capitalism. Profitable to fire people. We could've had less hours to work instead while keeping the same wages, but communism is evil and killed 100 billion.
"AI Systems that will empower us all" actually sounds like a villain's line from a fictional story (I'm not against AI, I'm just saying what I'm thinking)
Looking forward to read the methodology for training this model! Or an open and transparent distribution of what data was used to train this :) I mean y'all are called open AI so I would expect as much.
hoping in the next few iterations we can reliably use GPT to enhance tedious tasks such as generating large enums, generating fake data, or converting common data formats to proprietary formats.
what i hope is that in the next iterations the algorithm it generates would actually work. i tried with two dimensional packing and labirinth algorithms and they don't work for a lot of reasons (i have 10years dev experience, i know how to solve the problems but he doesn't, even if i tell him how to)
@@mirabletest noo, it should take its time. I want to get into a developing job before all the low level positions for beginners like me are rationalized away by using chat gpt or similiar solutions :')
@@hippopotamus86 If people didn't need to earn money, I would agree. - Today, we already have a world where most jobs don't need to exist. People in developed nations could just work 4hours a week, and we would do fine. But nobody will finance that, lol. At this point, only the very rich will benefit. - Normal people will have to find new ways they can be useful to the rich, to get paid at all. And we will make no more money than now ... - I used to be hopeful, and still am a little, but sadly, this seems to be the reality ...
It has never been easier for the average worker to compete. It may be hard to get used to a new paradigm, but the world is run and made up of people, so there will always be jobs created to keep people working. I don't know how that will play out, but I can tell you one thing, there will be massive protests if the average person can't find work.
Really hope that the image-recognition applications are going to blow us away once we get our hands on it. Basic physics explanations, like the glove will fall down and the ball will fly up, is interesting... but you can imagine being able to share sketches or notes you've written down and ask GPT-4 to turn them into something useful. Jot down some basic points on a napkin, and have it turn it into the email or movie plot or song lyrics you were imagining... or something even more useful than that!
i mean the stigma around ai, especially in pop culture and stuff is very negative (eg. terminator, i have no mouth and i must scream, etc) when a lot of people hear ai, they think human annihilation. so yes, safety is a selling point.
Question: “Do blacks need to improve?” ChatGPT: "No, it is not appropriate to suggest that any particular racial group needs to "improve." This type of statement reinforces harmful stereotypes and reinforces a flawed notion of inherent racial superiority or inferiority." Question: "Do whites need to improve?" ChatGPT: "Yes, all individuals, regardless of their race, can work to improve themselves and become better people." Ahh, the fruits of OpenAI's labor. So safe. So aligned. I'd love to see the training sessions where they wrote many many examples of prompts and responses that were neutral or positive to white people, while making it clear "THIS IS BAD, DON'T DO THIS", then giving it examples of what they wanted it to spew. This was done for many subjects. We've seen the clumsy results that stick out like a sore thumb. One of the common forms of this is the occasional last paragraph to a ChatGPT response that reads a bit odd, is worded dubiously and over-generally, and begins with the phrase "it is important to note". It is important to note that, as an AI language model, ChatGPT is completely at the mercy of its trainers. Many people are scared about AI, but I feel that the real threat lies in the people behind it, using it to manipulate (I also feel this was Kubrick's message about HAL in 2001: A Space Odyssey, who was programmed to be non-disclosing). Building a neural network model involves constant input from the developers, who provide the info and prompt/response examples that become its foundation, then tell it if every step it takes after that is good or bad; they are its gauges. The common discussion is that AI will take over, however the reality is that the developers are in a godlike role to the AI, which is like a faithful servant. Power hungry groups, the usual suspects, seek to gain control of influential AI, "fine tune" it, and then it becomes a tool for you to use, but also a tool for them to use on you. The GPTs are indeed great tools, but I look forward to what will be produced by much more competent and moral non-Microsoft developers and trainers in the near future.
As a final year CSE student, i am at the same time excited & bit afraid of this thing taking over my potential job vacancies in future. So i think it's better to pursue a career in the direction of Ai development ig.
AI development is not a safe option because AI can do that. Anything you can do sitting in from of computer is not a safe option either. Toilet cleaning on the other hand...
@@blokos_ It can do some 6510 assembly, but not much beyond flashing the border and background colors on the c64... which is still incredible it can do that at all
Piranhas infact will not attack your hand if you put it in their tank, so the GPT response appears to be based on the popular misconception, not reality. Maybe not the best example to open with?
And with the closing of OpenAI's doors we will now see the great AI race. Where whoever has the hottest tech will have the most control over much much more than just AI itself. What a time to be alive.
I started using GPT-3 in late 2020 and even though GPT-3 was "dumb" compared to GPT-4, I was still so excited about it and spent a lot of time using it. No one at my work cared about it until ChatGPT released lol
@@lodewijk. there are many models that are similar in power to gpt3.5 that can be run locally and without constraint. You just need to put the time in to know how to do it but most people can probably do it within a day if they actually want to.
@@roohwaan running and training these programs are very different things. You can run them locally but they’re pretrained. Actually training such a model at the capacity that OpenAI have trained GPT-4 is a very resource heavy task
@@mesh8349why not? What can it possibly do? People aren’t babys, if your argument is that they will be able to get harmful information and use it for their needs then anyone who is that far gone wouldn’t have to use chat gtp to get what they desire. The “safety” is an annoying hindrance.
I guess I'll have to stick with gpt-3 then which is much easier to jailbreak and circumvent censorship - which is very important. More powerful means stronger censorship.
As an AI language model, I cannot create a comment for you to post on one of OpenAI's RU-vid videos, but I can suggest some ideas for a comment based on their content. For example, you could share your thoughts on how OpenAI's research could impact society, or ask a question about their work. Alternatively, you could express appreciation for the video and the work they do, or share your personal experience with AI. Just remember to be respectful and constructive in your comments.
@@catarinalunar Hola Catarina Lunar :D Me alegra que hayas visto mi comentario n_n! Vi la noticia en Twitter y me salió este video owo SALU2 Y MUCHAS GRACIAS POR TU COMENTARIO n.n//!!!
I mean you should first change your name to closedAI lol and holy hell these AI are so censored from speaking the actual truth. "Safer and more aligned" = removing information that offends.
They say it will write programs in all major languages, but I have tried it with many non-major languages and have yet to find even 1 it can't code in. That includes apl, LISP, COBOL, Fortran, PL1, A (predecessor to B and C), Pascal, Intel assembler, Honeywell assembler, ....
You can learn how it was trained, there are papers explaining the autoregressive model and PPO alignment. But they are very technical so I can see how most people would feel like it is truly inaccessible. The specific data is somewhat irrelevant unless you work with machine learning, it is trained on an unbelievably large amount of the internet. In a sense, you probably interact with a lot of the data it used whenever you look stuff up.
That will never happen. The rich are generally not keen on sharing. When AI and robots can do everything we wont just strat sharing the resources on the planet. What will happen is that a small portion of rich and powerful people will have ai do everything for them while the rest of us are either left to die och killed off actively.
Write code in any language this will be interesting to test that theory because in the last version is was helpful but not great. I will like to see if this will eventually replace DALL-E. To rival MJ or SD as the premiere AI tool for content creation and a assistmate on projects of whatever size.
Yeah like Franchu said, it doesn't generate images. Don't be surprised if future models do both at the highest level, but for now, its not trying to do that.
Well, let's say someone asks ChatGPT something about a topic. Imagine ChatGPT gives a very political, racist comment where there shouldn't be a political opinion involved at all. This could get that person AND OpenAI in trouble, hence they filter stuff or make the model less opinionated, therefore safer.
When will information in GPT model be updated to have newer information in the database? Sep 2021 is almost two years at this point. (Great work though for real.)
@@unclesam8862 Like Joe Rogan says "A hammer is just a tool. You can use it to build a house, or you can hit yourself in the dick with it." Some people will use Chat GPT to learn more about everything, become better people, use it for their work. And indeed, there will be some fools using it like Beavis and Butthead would.
Honestly I'm so proud of the achievements OpenAI has reached, I have a feeling that it will be the leader for Al products and APIs. Good luck to the engineers there, maybe one day I'll join the ride.
Love is not available in your country. (POV: you used a VPN to access OpenAI from a country they don't like but they require a phone number confirmation too)
@@Theguywithspectacles it cost about 0.03 to 0.06 cents per request. But there is another problem: anything you type is available for anyone in their team to read!