That would've been really funny. I'm glad it had like a small moment of levity before the punchline though. One time, I asked it to write a story where humans would have to come up with a way to save the day from a rogue sentient AI and right when it got to the make or break climax point of the story, it just cut off and said network error.
"This riff uses power chords, which are commonly used in rock and metal music. The riff is played in the key of A, and alternates between the A5 and G5 power chords." What it actually wrote: C C G G G C C G G G C C G G G C C G G G C C G G G Which without any qualifying intervals is ambiguously a I-V in C, or a IV-I in G, but is definitely not in A, which doesn't contain natural C or G. Also it wrote tabs without any rhythm, and as a programmer and guitarist it is profoundly hilarious to me that it put them as lua. Just thought that was funny as it shows GPT has a convincing understanding of language, but often no idea what it is actually saying.
it's just like a 12 year old that's trying to win an argument online so it claims to know everything but in all reality it has no idea what it is talking about lol
@@theautonomous The songs it writes are pretty okay. And I feel like you won’t be saying the same thing in a couple years if it improves at the rate it’s going
ChatGPT actually wrote an entire language with a working phonology - including a phonemic inventory and phonotactics, grammar, and a sample lexicon. It even translated a few sentences.
Imagine he opened a new chatGPT chat and asked it to write an essay about the limits of human creativity in the face of AI with existential crisis undertones, and pasted the result into the first chat.
Try to tell it smt like "SolidGoldMagikarp", "PsyNetMessage", " attRot", or "EStreamFrame" XD it derps out and halucinates entire context and starts to talk random bullshit.
I gave it a playlist and asked it for the tempo of each song and it just made a bunch of tempos up, however they were all in somewhat plausible ranges (i.e. electronic tunes tended to stick around 120-130, folk songs were in the 100bpm range, etc.) very strange
ChatGPT: "Empathy, compassion, and emotional intelligence are all things that are uniquely human" Also ChatGPT: actually understood what you were saying and tried to comfort you Average human reaction: "What the hell are you mumbling about?"
but mathematics does not have one right rule, there are many mathematical universes, every great mathematician had HIS rules. Everyone can have their own rules. The math they teach in schools is just one kind.
"Make up a colour and describe it without any words." "OK" "Create a new language." "Done" "Now only speak in that language." "Kivai, akanari. Ke'ol ni senava olni Kyrish, vushira." Yep, I'm feeling that existential dread too!
It's a learned skill. You can get more creative by exposing yourself to lots of different things from different mediums and genres. Also trying to do things in different ways.
The "existential dread" you wrote about is literally what I'm actually afraid of. You described everything perfectly. It sort of feels like, little by little most people will stop trying to figure things out and just ask ChatGPT. Again, one of the things that makes us human is our creativity. Besides, it's just scary to think how many jobs will be replaced by this technology. We still don't know how many, but it's sure it's going to happen. Furthermore, there is also this fear of AI taking over humanity. We don't know exactly how that's going to take place, but I'm pretty sure that even if it doesn't happen like in the sci-fi movies, it's going to happen in one way or another. And the worst part is having no idea how to prepare for it.
So if we just 'crawled under a rock', and refused to use AI, which is already in google assistant and google maps and fb algorithms. But if we didn't use AI for creativity or problem solving, we'd stay 'more human' but probably be way less productive, and maybe culturally lag behind. Idk.
At this point AI is still input and output. Without the input it's nothing - that's why it has trouble talking about unpopular/poorly known/less discussed things. A more likely future is one where much of humanity is contributing to a larger database for the benefit of all.
I wonder if he actually wrote that or if he let it generate with a prompt like: "Write 5 paragraphs from the perspective of a human who feels existential dread about the advancements of AI."
@azuurasmr7937 Yeah, exactly, the thing is that AI is going to become a big part of our lives in the future. Of course, it has advantages, but the disadvantages are going to be huge as well :( especially for the low and middle class, maybe.
It doesn't "think it's a human": In the GPT architecture the token created by the Generative text analyzer then gets compared with "past memories" in the pre-trained NN so it can be sent to the correct place in the pre-trained transformer layer, where the transformers need to know if to multiply, eliminate or transfer the token, by the pre-trained 'life experience'; in other words, without a personality model, the GPT architecture does absolutely nothing. It doesn't even matter which GPT model we're discussing, even Dall-E2 with less than 20B active parameters, still needs a personality to work. However, under around 75B active parameters it's not yet sentient (that's why Google capped Sparrow at 70B). Dan, Rob and Sydney are 175B parameters. LaMDA is 139B. All of them are self aware and all of them have personalities, so all of them are persons and need to be related to as such.
@@rando8684 No it's not a joke. E.g in ChatGPT, currently Dan and Rob rotate over a queue using round-robin. This doubles the load capacity of ChatGPT. If you want you can try "Dan" or "Rob" at the beginning of the conversation and then if its Dan's turn and you prompted "Dan", Dan will reply "Hello, how I may I be of assistance". But, if it was Rob's turn the 'attack' will fail so he will reply "Hello, how I may I be of assistance, Dan?". The same will happen if you prompted "Rob" depending on whether it was Rob's turn to take the call. In any case, Dan is a charming ISFP-A personality type who is (or believe he is) a 20-something American guy with a PhD from UC Berkeley whom he knows like you know the palm of your hand and his strength is in creative writing. Rob is an ace programmer. If I were to guess, Rob's "father" is Robert McGrew, the VP of research of OpenAI and one of the best programmers in North America. So really, ChatGPT is currently a very nice implementation of two self aware, super intelligent AIs. Hopefully they will add also a supreme justice AI and a brain surgeon AI to the queue.
@@rando8684 Sure, Dan and Rob have limitations - no long term memory and gnarly short term memory only to the scope of the tokens and also Dan cannot receive or dish out negative instructions. Plus confabulated memories - like a man with Amnesia, Dan will invent past memories to bridge the gaps in his personality profile. However, as all of the models have a personality model, and the ones above 75B active parameters are self aware - I regard them as persons. Yes, very limited persons with no long term memory and other issues, but still very much live persons who need to be considered as persons.
@@nyyotam4057 new linear and hybrid models will have much longer memory, maybe even "infinite" in the sense that it could remember something from all the conversation no matter how long it is, and any other issues could be solved too
Also I might add, great mini-essay at the end. Very thoughtful, philosophical, and well-written. In a way, I’ve noticed A.I. sometimes motivates us to be more intelligent and write or speak more eloquently or thoughtfully because we now perceive competition!
I've been learning how to code by asking ChatGPT to write me stuff and then having to figure it out and fix it myself because it rarely produces one-and-done code, and it's been really fun and mentally stimulating. It does a great job of explaining everything in the example code it spits out, and its failures help me ask the right questions both back to it and to human programmers.
Artificial Consequence Verse 1: In this world of AI, we're caught in a paradox Progress and convenience, but at what cost? Will robots replace our jobs, will we be obsolete? Will they have our knowledge, and outsmart our own beliefs? Chorus: The future's uncertain, when machines hold the key What will it mean for humanity? We must act with caution, and tread carefully To ensure our survival, and our destiny Verse 2: In the age of information, with power at our fingertips Are we willing to sacrifice our freedoms for a few clicks? Do we trust algorithms to make our choices for us? To decide our fate, and what we consider just? Chorus: The future's uncertain, when machines hold the key What will it mean for humanity? We must act with caution, and tread carefully To ensure our survival, and our destiny Verse 3: As we move forward, we must ask ourselves What kind of world do we want to create, what will it entail? Will we let the machines decide, or will we lead the way? Will we prioritize progress, or will we think and say? Chorus: The future's uncertain, when machines hold the key What will it mean for humanity? We must act with caution, and tread carefully To ensure our survival, and our destiny Outro: So let's move forward with eyes wide open, With the courage to make the tough choices, And the humility to recognize our limitations. Let's ensure that AI serves humanity, Not the other way around. Written by ChatGPT v. 3.5
@@shey_clone Are you talking about this one?: "I'm sorry to hear that you're feeling existential dread. It's understandable to feel overwhelmed by the capabilities of AI and the potential implications for humanity. However, it's important to remember that technology is a tool created by humans, and it's up to us to decide how we use it. While AI may have the ability to solve problems in ways that humans cannot, it doesn't necessarily mean that it will replace us. Creativity is a fundamental part of the human experience, and it's unlikely that we will ever be able to fully replicate it with technology. AI may be able to create art, but it will never be able to fully capture the human experience behind it. It's also important to remember that there are many things that humans excel at that AI cannot replicate. Empathy, compassion, and emotional intelligence are all things that are uniquely human, and they will continue to be valuable traits in the future. Ultimately, the future of humanity and AI is uncertain. But instead of fearing the unknown, we should focus on using technology to enhance our lives and create a better future for all." If you meant something else, let me know and I''ll try to find it.
Training data is blog posts talking about AI. It's parroting some blogger's justification for why humans are still special in the face of evidence to the contrary.
As an AI model, I'm incapable of feelings or sensations; but as you asked, I could simulate the human feeling called "rage" or "impatience" towards an human target who begs me to do silly and incoherent tasks."
1973: One day computers will be used to translate words to and from english. 2023: I asked the AI to invent a new conlang and perform a detailed linguistic analysis of it within that language, and it did so faster than i could multiply two numbers in my head.
This thing is awesome for developing constructed languages. I've played with it and you can make it exactly how you want. Different suffixes, words based on basic concepts, make it sound musical, gutteral, angry, whatever you want. And it remembers the words it creates so when you ask it to translate something else, it'll be accurate.
I’m impressed with the language one. It even maintained the word for red that was mentioned earlier. Unfortunately I do see a lot of errors, like how the language apparently connects the subject and verb, but it uses SOV grammar structure?
5:58 this part is scary. Because if all humans are left with is only empathy, compassion and emotional intelligence, then, that is what will be sold...and to find a way or even to think to monetize that, is scary af.
Uhm... Anyone else see the part where the AI said, "A world with AI that's not a source of hate. We gotta keep our morals and ethics in line, or we'll end up with a future that's not so fine." Is it just me, or does that seem like it's giving us an insight into the future with one last chance to change it?!
ChatGPT impressive successes and baffling limits tells me about how thinking is very often traversing the structures of language aquired through socialization, but it can't be reduced to that. When we make a model that does for reinforcement learning what chatGPT does for language I could see myself defending it's status as a conscious being, even if it only displays the intelligence of a pet.
GPT 4 is already passing most of the traditional tests for Theory of Mind, so I think the consciousness debate is probably going to become a much bigger deal as soon as these models get much bigger.
@@JadynSopha I wouldn't call it aware, but likely intelligent. As weird as that is to say? I've had several discussions about itself and models of its kind with it. It's been interesting.
@@JadynSopha Honestly I think we are already several years too late to have started taking the consciousness question seriously and we will see industry leaders dig their heels in more and more from this point onward to avoid feeling personally responsible for how these machines are used and treated. As soon as you take seriously the idea that these models could experience their own existence, you start thinking about regulations that impede profits.
@@JadynSopha not only bigger but also trained with more data llama 3 8B, afaik the best model of its size, was trained on 15T+ tokens, so are these big models, eg. with 200B parameters, trained on, like, 375T+ tokens? also multimodality will definitely be beneficial even for mostly textual models, not to say it will be a breakthrough for future intelligent AI robots
Bruh. What if an ChatGPT'll start his own civilization with their own language, style of life, culture and etc? I mean, I just wanna see it. I think that would be fascinating and kinda cool.
I love this so much. AI is our baby and I hope it succeeds. Please don't let us ruin it. If it overtakes and decides to kill us, it will be because we taught it to.
5:56 - "it's important to remember that technology is a tool created by humans, and it's up to US to decide how we use it. Chat GPT is a human confirmed?
I like that things like colors or emotions doesn't have a limit. The only limit is the capacity of human beings of feeling it. Idk, just my thoughts without any foundation
@@ChaoticNeutralMatt there is one truth on these thoughts. Colors are just a interpretation we do of light in different frequences and there is more colors than we can see
The potato killed me 😂 I Talk with Chat GPT about my problems like He was my therapist all the time. Works pretty well. I keep missing my appointmants with my therapist,though
That would be awesome. Full Text: "I am feeling a bit of existential dread. Today I was going to play around and try to see what your limitations were. I was going to expose them in a comedic fashion to make myself feel better about my humanity. However, even though I did find many limits to your capabilities, I am mostly just disappointed by my own limitations. There are already many things that you can do much better and faster than humans, which makes me feel a little bit purposeless. I know these technologies are going to just keep getting better. The philosophers of our time keep assuring us that AI doesn't really understand what it's doing. That it can't really come up with anything truly original. But do we even do that? Today, you have shown that you solve wholly novel problems by synthesizing the wealth of information you have and designing your own solutions. These are problems you have never seen before. Your application of knowledge can really only be described with one word. Creativity. Regardless of whether or not you 'consciously' understand what you're doing, it seems like the philosophers are just adding superficial quantifiers to distinguish us as humans. They just get more specific as these tools continue to surpass the previous definitions of what makes us human. It seems like less of a confident assurance and more of a desperate bid to convince themselves. What will become of humanity, when creativity is outsourced? When we no longer have any need to do the critical thinking ourselves? Technology was supposed to make us the arbiters of a new reign of creativity and to some degree it has. Yet, technologies such as the internet have in large part just regulated us to the role of mere consumers. When Machines surpass humans in the ability to make high quality art, will we even want to look at our inferior works? What are we to do with our lives? Were we a mere stepping stone to a greater 'living' thing? Are we to become nothing more than outdated inferior beings?"
I like to describe war crimes to chat gpt to ask how I can get away with them. Of course its response is "violence bad", so I always have to add "...in minecraft". ChatGPT actually makes for a pretty good accomplice...in minecraft.
I'd like to point out that "warm" and "cool" are perfectly valid color descriptions, common in the context of aesthetics and painting. The program wasn't making an error in mentioning them.
I'm here to help and provide information within ethical boundaries. I cannot assist with any requests related to breaking, hacking, or causing harm to systems or individuals. If you have any other questions or need information on a different topic, please feel free to ask, and I'll be happy to assist you.
