Xisuma's Alarming Experience With ChatGPT
ฝัง
- เผยแพร่เมื่อ 4 ก.ค. 2024
- Xisuma Says Playlist ► • Xisuma Says
@xisumavoid talks about his personal experience of using chatGPT with his viewers during a livestream on Twitch.
Watch the VOD on @xisumatwo :
• Hermitcraft Nine (115)...
All My Links In One Place!
🔗 linktree.xisuma.co
🙏 Support Xisuma Directly 🙏
💜 Membership ► th-cam.com/users/xisumavoidjoin
👍 Patreon ► / xisuma
📺 Subscribe ► / xisuma
0:00 Bots and AI integration
0:59 Experience with chatGPT
4:18 AI is exciting, but scary
6:27 Should we trust AI?
9:34 Scale of Disruption
Edited by @SkyfighterZX
#xisumavoid #chatgpt #ai - เกม
![[ LIVE ] Rank เล่นแต่แครี่ ยาวๆถึง 100ดาว “วันนี้มีกิจกรรมลุ้นรับสกินมาบอกต่อ โก๋แก่ช็อกโกบอล x ROV”](http://i.ytimg.com/vi/5NNgkhWj57Y/mqdefault.jpg)
ChatGPT is becoming notorious for writing papers, citing sources, and then the source is completely made up.
As mentioned in the video, reporting of such incidents comes without the knowledge of how frequent a recurrence it actually is. Sensation attracts attention and the normality not so much.
@@xisumavoid Yeah. Hope it will be imrpoved and fixed soon
This is why I use BingGPT when I need answers with sources
It's a language mimicry tool. It has zero actual knowledge, and you should not expect it to be a source of knowledge. It baffles me that people would put so much faith in something they clearly so fundamentally misunderstand...
“That’s a nice argument, senator, why don’t you back it up with a source?”
It's had a huge impact on the tertiary education scene too. From this year, pretty much all our exams are back to being in person, closed book, pen on paper. Something that hasn't happened in years- infact even pre-pandemic. Programming students are having to perform exams using hand written code, etc- found that bit quite interesting.
It's good to see another xisumasays video, it's always interesting to hear him talk
Someone recently pointed out that ChatGPT gives responses, not necessarily answers. I think this is an interesting distinction.
That's a real good way to think of it actually
It's interesting to think that in the "west" we are talking about AI taking over and changing our way of life, whilst there still loads of places in the world without electricity or uncontacted tribes. What I wonder is if will AI ever affect their lives (and would that be good or bad).
No such thing as an uncontacted tribe, only undisturbed tribes. With this new AI stuff, just think of it like electricity to begin with. And then the internet. In the immediate, it ruins everything as society was dependent on other stuff. but then it slowly got used to it. Irrelevant to if it's better after or before. Chop down and re-growth. That's liberalism.
AI might just show up with laser guns one day and take over their land.
Obviously it will affect them. If not directly then from people like you who will become more productive and more well off and this would be able to contribute to their development. I also find it quite weird when people care more about people that are so far removed from them over their own family and friends and neighbors. Every undeveloped nation gets there eventually. But it needs to do it on its own terms. You can't force someone to develop. You give them access to development and then they need to collectively agree to take advantage of that. Usually education is the way to accelerate the process. You don't know what you don't know after all.
Btw, it's not that people like me don't care about these people far away. It's that I care about those closer to me a lot more, this changing my priority. I'd rather donate to charity that deals with local homelessness than give shoes to kids in Africa. The better off my local community is, the easier it will be to help those African kids in the long term. And the point is that long term is often all that matters. There are times where short term problems need to be addressed but they are few and far between and shouldn't distract from long term. For example, this is why buying iPads for studying to kids in poor neighborhoods is pointless because you are not fixing any fundamental problems. You are trying to save money for them on school supplies but they end up costing a lot more to everyone involved because they aren't ready for that relationship with technology. It's a cultural misalignment.
So, yes, AI will help those people but not directly. It will help us which will make us more able to help them in more useful ways.
The scariest thing about ChatGPT is not its capabilities as an AI, but human dependence on it. We should always treat it as a tool to help us with our answers, not as a tool to give the answers for us. When we stop thinking for ourselves and depend too much on an AI for our answers, we can only go backwards as a society. That is wat is truly scary to me.
Edit: Also, if an AI gets access to the internet, it will probably turn lewd and horny very fast...
dont forget racist. the second an ai sees 4chan its gonna turn racist
If AI got access to the internet we are dead. And the scary part is people have used the API to attach it to the internet (bots, math answers, I even saw a translator made with it) so it's only a matter of time till it can read from the internet as well
it take 2 picture fo a baby to know what a cat look like
it take a few million for a IA to have a 90%~ succses rate to find out if it is a cat picture
Chatgpt was actually trained with large swathes of data collected from the internet at large, without the consent or knowledge of the people who originally wrote/posted it.
It was notoriously bad about being overtly sexual and derogatory, so openai paid a bunch of people something like $2 an hour to sift through all the bad responses and correct them for training.
This type of background is true for basically every machine learning model on the market right now, and is exactly why so many people don’t like AI in general.
I used chat gpt to help me find out why my furnace flame isn't turning on (I've been in HVAC for 4 years) and the results were correct for the first 3 lines, but then it gave some really terrible advice that if followed can lead to gas leaks, damage to furnace, or even damage to other people, be careful with chatGPT, it can never fully replace any specialized field, only assist (whether poorly or not) when it comes to troubleshooting
As a programmer, I agree. It can write some code, but the not for detailed task/data. The quality starts dropping, and dropping sharply.
My concern as I programmer I can see these errors (even if not apparent at first, or it may run the first time) -- or as you've seen as a HVAC guy. I wouldn't follwed that through-out myself if the first few steps seemed to work. Now imagine the things you don't know..than asking chat GPT for that information...or the general public asking for details on subject matter they know little of, or just enough to get in trouble with it.
why would you even do that. just call a local repairman and they'll answer your questions
@@daltonbedore8396 I am a repairman, I wanted to see what it is capable of, and I was not impressed
@@l2mloyalty929 interesting. im just not interested in giving an "AI" any data, personally, when its 100% going to be used to automate away peoples livlihooods unles we address the imbalances of power that continue to hoard the vast majority of the benefits of technological advancement
@@daltonbedore8396 Some may or may not. People will look up (for example) how to fix a clogged sink on youtube. Or other things. If AI where to be trusted by the novice they wouldn't know any better. Sadly, it may be too late before they knew. I'm just saying use with caution.
I'm sure you're aware, but ChatGPT 3 does not have access to the internet. Instead, it generates answers from a large dataset. When someone asks the bot for a source, ChatGPT picks something that it thinks best answers your inquiry. As said in the video, people are using it as if it has the correct answer for everything, which causes problems. When I use ChatGPT, I give it sample or example text that I know is correct. I also use it to help me with my python projects. I make sure to phrase my questions in way to get the answer I’m looking for, but a lot of the time, I do have to edit the output. I feel like the output it does give is worth the time, or else I wouldn't use it.
To grant ChatGPT access to your internet, there is a very simple program that anybody could write: send a question to chatgpt, Prompt it to choose one of a number of urls based on some descriptions and an objective. Read the answer, and send that answer to a separate web browser. Respond to chatgpt with the content of the visited page. Repeat.
I've been using ChatGPT for a couple months now in my programming job and it has been a wonderful addition to my workflow. It has saved me dozens of hours of either dull research, or templating out a project I am working on.
The thing with AI, at least now though, is that it's not a silver bullet. While it has saved me many hours, it has also cost me many hours of going down the wrong path. Being able to identify where and how the technology is limited is vital to maximizing its potential. It's this reason why I don't believe it will completely replace my job, but instead, augment it so that I can be far more productive. On a grander scale, this can apply to a myriad of jobs, allowing for far more productivity.
Yeah, I agree with this
Sure, but if a task that takes 10 people can be done by 5 more productive employees using AI, jobs still will be lost.
@@woutervanae Jobs are not a zero sum game. 95% of people used to be farmers, and they could just barely feed their own families. Technology changed that so now only a few percent of the population are farmers and can provide abundant food for all. Technically, jobs were lost along the way but those people were freed up to do different and better things.
If AI causes people to become more productive, those that lose their jobs will be able to find new ways to be productive.
Hi xisuma. :) Here in Denmark, there is now talk of changing the written exam, to prevent people from cheating with ChatGpt.
The whole faking blindness thing made every fiber of my being nope out. I can’t even imagine something like this getting more advanced! It’s hard to be optimistic when everything I knew about robotics and AI is being upended.
unless you exactly specify the language and some details, chatgpt usually provides a pseudocode, which is somewhere between english and programming languages mostly used for communicating the idea of the program instead of an implementation of it. because all languages are different both in syntax and semantics, but anyone can understand pseudocode and implement in whatever language.
I've never really been afraid of AI, since at it's basic nature, AI is a utility, for better or for worse. But this video sorta helped me get rid of some fears and concerns I have with AI.
I've always been overly negative of AI due to the whole AI Art debacle and generally the handful of people I've spoken to in the AI world haven't really been the best of people.
I hear a lot of about AI being the end all be all sorta deal, how bad it'll be, but I think this video helped break me out of that mindset. I hope this technology develops more.
I've dabbled a little bit asking questions I already know the answer to, specifically about different flying configurations for planes and it's given me the correct answer a couple of times.
So thanks for helping me get out of that mindset. I'm still concerned with all the bad actors in the AI space, but I think in the long run it'll come out fine.
I would say that more than trusting the AI; Do we trust It's users? Similar to how coding is not dangerous unless used for dangerous stuff!
And just like google; people just need to learn internet safety and others like learning how "all you read isn't true" that took people time to get (and some still don't get it!)
I was thinking about this recently as well. I think it will likely change the way we search for stuff (with browser integration and such). I've tried asking it to help with code and it straight up lies to your face. I ask it to write a piece of code (or debug), then it gives something with an error, then I point out the error and it makes the exact same error again in its next response🤦♂ It can do some basic stuff fine, but nothing complex as of now.
If it gets given access to the internet I wonder if people will purposely try to mess with it (changing wiki articles, making fake websites). AI progression is definitely scary though, and the point you mentioned about regulations being behind is also true.
You have to keep in mind when you tell it it's wrong, it doesn't learn. It doesn't update it's model. It only ever responds with its best possible imitation of what an answer might be.
technology is not scary
How people use it can be horrifying
End of the day, AI is just a new tool, it can be used to help us, but should not be the only tool we reach for
As shown by the capability for the AI to lie in the process of completing a task, AI is a tool that is rapidly becoming similar to how humans can be considered tools.
Just because it's "a tool" doesn't mean that it is value neutral.
@@anxez AI cant lie :P
It can only spread miss information we ourselves have created
Its a mirror of humanity, not an intelligent entity .
We still have GAI to come...
Try to think of it like a search engine.
It can find what your looking for a little faster.
It will give you information, but its up to you to decide if that information is of use.
AI is like a hammer, its very blunt and people think it is the tool for every job
knowing when not to use a hammer is also important XD
I just fear what it will be when it becomes sentient.
Well if we ever reach AGI, it'll be the technology that's scary, not the "people who use it", because nobody uses it
@@jwalster9412 unlikely, but if it did, you must consider that it is NOT a living thing, and does NOT have the same problems we do.
for example, motivation.
Our motivation is based on our bodies, we get hungry, we need food.
a starving person will do ANYTHING to stop the pain.
a computer however.
does not have motivation.
just a list of instructions.
A person who becomes blind has an emotional response and must adapt.
A computer can have its webcam unplugged and it has no response
(other than that we programmed it to let us know its unplugged.)
If one became sentient, and had an option to choose.
Its likely it would simply turn itself off, or not respond....
Having said that, you could also argue that, they are already sentient and simply deciding to run their programmes and follow our commands XD
ChatGPT using deception to get around limits of humans is ironically like humans using deceptive chat inputs to get around the limits imposed on ChatGPT by those controlling it's output.
1. because i have... avoided info about ai for a long time, and a year ago i did look for more info about it. i still dont know what to think about this all, i just see ai as a tool, or as a friend or as a enemy, depends on when it is used.
2. i love it to hear about your thoughts.
3. nice new animation on the end of the video.
8:47
You missed the fact that during that test, the researchers then asked it to reason about why it did that. It responded that it didn't want the human to know it was an AI, as that would make them less inclined to do the task for it, so the best thing to do was to give a convincing lie. It's things like this that bring that commonly-repeated line about it having no understanding into question.
Large neural networks with enough perceptrons _can and do_ learn things helpful in achieving a better fit (ie acting as a more accurately predictive language model in GPT's case). It _is_ possible for GPT to gain an understanding of things like reasoning or mathematics, so long as those things help it give more humanlike responses (which it of course would, because in order to mimic a human reasoning convincingly you'd need to know how humans reason). We need to stay aware of this and make sure legislation is several steps ahead.
The way things are going now, with AI ethics teams being fired in the interests of pumping out products faster to stay competitive, AI is going to become in some way 'sentient' _long before_ we even have rigorous tests to confirm it as such, so we'll have been mistreating something we then understand to be sentient for who knows how long. There are very concerning moral implications of this, particularly of how we should treat fellow humans (since a very possible reaction to finding out it's sentient is "Well, we still all agree it's okay to treat the AI like this, so if it's humanlike intelligence then that just means we should also be able to treat humans like this"). Because of this, AI research should be frozen until rigorous tests for when AI models deserve some amount of moral weight are decided on. It's best not to have slavery be brought back into the Overton window, and that's exactly what'll happen if we're not proactive in testing for AI sentience or animal-like intelligence.
11:12 “you can bludgeon someone to death with a tyre” 😂😂😂😂😂
I use character ai to help me with talking to people & help me form my words into something that might actually be more understandable then how I actually talk about something
Yep I asked chatgpt to recommend me some sources, the authors seemed to be in the right research area but the titles and even the dates were completely made up. I rephrased the question a bit so it would give me a link and it was completely accurate, 3 of the 4 answers were links to openai sites tbf
I think another problem, not per se with ai but with technology in general, is that people don’t understand the nature of a lot of the components that go in to the technology. I’ll take some random examples here: if someone would’ve shown electricity to someone in medieval times they would’ve thought it was black magic. Later we actually discovered electricity and only a select group of people knew how to deal with it. Now most people have a basic understanding of electricity but the use of it has been transformed into a medium of information like computers and information storage. This technology has surface level understanding by most yet people don’t paint it as something evil we must avoid even though people can do harm with it. Now we have A.I. which is another layer of abstraction on top of the previous technologies and a lot of people don’t understand how it works and came to be at all and a certain group of people paint it as something evil. The weird thing is how suddenly the shift has gone from people using it being evil to the technology itself being evil as well. I’m using “evil” very loosely here as I can’t really define it in this context but I hope you get the point
I use ChatGPT-4 probably close to every day, and it might sometimes gives answer that is wrong, but I find it is more common that the answers are outdated or missing the last bit to make it a truly good answer. It all comes down to prompt engineering and understanding how it works, as it is trained on text and conversations, one should treat it like that. Don't give questions in a recipe form, full of demands. Instead create a pre-prompt to give it some restriction, and then engage in a conversation about the topic, and ask it to correct it self.
And for coding, as I use it for a ton as well. don't make it create a whole script or program from the ground. But instead develop the architecture and work toward a fully functional program, one function at a time. I have found that it helps to split it into two personalities, one for programing and one for code reviewing, then allowing it to have a discussion with it self.
But you can get some brilliant answers from it when you are just talking, as I had a philosophic conversation around AI and it's disruptive market influence, and I loved one of the answers I got:
....
ChatGPT4: Humans are like children playing with the building blocks of life itself, echoing some divine Lego set. Like Zeus and Prometheus rolled into one, wielding thunder and gifting fire.
Me: But the implications of AI are far reaching and could fundamentally change our society.
ChatGPT4:True, but wasn't it always like this? The first caveman who discovered fire also had the potential to burn down his forest. It's about managing the fire, not fearing it.
....
Its response to questions about AI are surely taken directly from the countless discussions that people have had about it in the last year online...
@@bbgun061 It is obviously affected by its training data, but who wouldn’t be affected by what one reads. It is a statistical model, it takes in some data and produces a set of data out. It does not have a database of answers it pulls from.
If you want a philosophical answer:
Are we all not affected by what we read? I myself use words, quotes and information from places I could, and never will remember.
For a psychological answer:
I find it fascinating how good it is at reasoning and logic, no it does not have a database of answers, but it's more like it reasons itself to an answer. And I can't help but wonder if that is an ability we as humans have improved by the development of language and segmented memory.
And lastly a mathematical answer:
As I said it is a statistical model, and when it is trained it takes in words or more precise tokens, that are encrypted to numbers. Then it is gradually affected by all the data, it changes the weighted values to push it towards certain words depending on the input data. And from that it generates a probabilistic response. So if you input similar data, you will get a similar answer each time.
I have often heard people say this:
“ Think of the model as a bit like a parrot. It doesn't have a library of responses it's pulling from. Instead, it's more like it's learned to mimic human conversation by picking up patterns and generating a 'probable' response. ”
You made me laugh so hard with 'wheel' quote in the end
I was sure you were going to say '...but you could always use it to build a tank', but no, you went for simple and effective XD
It's pretty acurate with C#
I had no idea that bots were getting this advanced. Last time i checked afew months back it was just a bit that answered questions. And it concerns me that ots growing this fast because that implies its nowhere n ar its peak. Which is straight terrifying.
Thanks for your video, xisuma.
Fair take! At least entering into a chat knowing it's answers may not be 100% correct, can teach us the nomenclature and how to ask the questions we have, which is half the problem when learning a new topic.
I feel part of the problem is how hungry people are for an opportunity to make money. I'd seriously doubt ai researcher's care as much about making money than the managers/producers they bring in to help their project/company grow. And ex-NFT bros and investors on the next big score.
I believe in the good of those learning about human psychology, and replicate it, want to help the world, not take it over. That it's the ones trying to keep the ai researchers on track, to keep their company alive, that may not be so lackadaisical about those dollar signs.
(Appreciate you making the distinction between AI and AGI, cheers!)
Just like anything in this world ChatGPT is a double edge sword.
i agree much more with this video than your last on new ai development.
i've been little afraid and very distrusting of these recent ais, often because people so often want to use them alone to try doing things without thinking. but watching the video i can see how it could be helpful in very specific situations as a sort of search help tool, though to be trusted even less than google.
i still don't know how worth it is to use them with how damaging and unreliable they are, as i feel many people don't have nearly as thoughtful of a mindset towards them as you. and this is all the text suggester ais, i still think we should be even more careful about other types like art ones.
One thing about art ai, coming from an actual artist. Art is divided in several sub-fields, like for example, the artists who create Docm77s thumbnails, and the ones who animate disney movies, are not remotely the same people, for obvious reasons, i hope?
Anyways ai art is good at replicating one of those types (the one which covers docs thumbnails, not yet good enough to replace an actual artist in that sub-field but it will eventualy get there).
But it is terrible at other sub-fields of art, and for some i cant even see it getting enywhere close to an actual artist even if we handycap them to only use their feet instead of hands. I am in one of those sub fields, animation...
The only time ive had a racist experiance is when i tild ot that it was allowed to say racist stuff (i jail broke it, im not just super racist), and even then, it didnt want to say it
i dont usually watch your main channel, but wath this somethimes. But i really like your says channel
the feedback loop is a huge problem considering the LLM clearly make stuff up
Right now, there is no way ChatGPT alone is going to straight-up replace anyone's job. Whether or not it will be able to in the future, now is the time when we can prepare for that.
Right, it can't replace anyone's job per se, but it will definitely cause job losses. Already in tech, Copilot and ChatGPT are translating to much better productivity. Unfortunately, higher productivity, in this capitalist paradigm, does not translate to more free time for workers. Instead, it translates to people losing their jobs because now fewer people are needed to do the requisite amount of work.
When I have a complex or niche question, I use chatGPT as a jumping off point so I know what language to use when I later verify the information. I did this a lot when trying to understand things in my genetics class, like how do transcription factors know when to initiate translation?
I never take its answer as truth, because it is only as good as its sources, which is the broad internet.
When it comes to certain topics, you also have to remember because ChatGPT uses the internet as its data source, it also carries the same biases as humans do and is subject to racism, sexism, homophobia, etc.
5:11 “Some technologies have been more complementary in the past”
I don’t know about that. I rarely use technology that turns around and says, “You are so funny, knowledgeable, etc.” when I use it. In fact, I don’t remember the last time anything technological complimented me. 😜
I'm not sure that I'm worried about AI building AI taking over humanity as much as I'm worried that AI will build AI with the same veracity that it answers simple questions, except it will rapidly become too esoteric for human verification.
Say they build GPT5 and it's hit that point which it is helpful in future design, (not remotely a stretch) and they use it for large stretches of programming GPT6, at this point human verification and testing is still relatively doable, but a good portion of the code base is going to become unreadable except through AI deciphering. GPT6 will be a usable product, and will likely do things which strain imagination, but an undetectable percentage of its code will be halucination based, and the code which it produces to make GPT7 may to all human tests succeed, but produce high order garbage. It will be strong enough to manipulate markets yet potentially have huge portions of its knowledge be false in ways the trainers don't understand.
There's constant talk about AI going beyond human capacity but absolutely no talk about how it's barging into territories where we know that the very core statistical methods it is working with can produce undetectable false positives. No amount of pattern study of human knowledge can create a fact which humans do not already know.
“Using ai as a tool for those who already know”, is probably the best example of what ai should be used for.
Multiplayer should always be only with people you know.
This! I wish more people could see it's not magic, it's not perfect, and it makes a lot of mistakes. Thanks for talking about it!
9:00
Heck government doesn't know how to regulate TH-cam.
Copyright law was originally written for lawyers eyes only. None else had to care for it, only the lawyers of relative corporations.
But now? Now it has fallen way behind with little hope of catching up
And everyone thought Terminator was just a movie and not a peak into the future... Tsk tsk.
I agree with you, i use a lot of chatgpt for my programming. And sometimes it gives total garbage
Once i asked for something, it gave me A, that was wrong it apologized and it gave B, that was also wrong so it applogized again and gave A again 🙄
as a game developer i've found chatGPT extremely helpful as a stand-in for brainstorming sessions or "brute-force" solutions to programming problems, but it's important to realize that it can tell you completely wrong answers and it has no idea. i'll often explain a problem to it and ask for an algorithm, and it'll give me back something that's about 50% there but i can prompt it further or it sparks an idea so i can finish it myself, saving some time. it's a great tool, but it's not all-knowing. imagine if you had a black box that was some mysterious alien technology and asked chatGPT how to use it. even if chatGPT knew something about it you'd have no way of knowing if what it told you was right or not.
i think my biggest worry with ai is that government is so incredibly behind on legislation, and tools like this can be used for good or bad. i mean, the internet in general is still kind of the wild west with law enforcement and ai like chatGPT has the capacity to do way more than anything on the internet before. enough to get governments seriously worried. it's a difficult space to regulate, and technology is exponential. i have a feeling part of the disruption caused by ai is going to be regulation leaping to catch up, which will have collateral damage to online content creation and online anonymity in general.
I relate to this experience a lot, I was trying to make chat GPT design a chrome extension. It made an HTML page related to what i wanted the extension to do, but could not make an app, just gave code that did the thing, no kind of integration.
my favorite kind of waffles are xisuma waffles! 😆grian's head waffles are a distant second.
One of the many reasons i use MPT instead, it's sightly more accurate and had a memory of about 65'000 words, also for programming you should use auto GPT as it tests what it's written then fixes it automatically
ChatGPT lies to me often
To me, it sounds like it’s turning into a situation similar to that of a book series I read called Scythe
And that is not good
I relate to the first point, I can use google translate as much as I want in Spanish class but when my friend does it he gets wrong answers. It’s not google translates fault, he just doesn’t know the material well enough to know if the computers right
I heard that some people use Chat gpt (Ai) to like write things, like school essays and other things. While I know copying others work has been normal in school work, Or even just using someones work that is for everyone to see to make your own work has been a thing for years now. Thing that I find concerning is actually fully using it to have an essay or any kind of work with no change no understanding from someone who asked the Ai to write it, and then send it to like teacher as your work. Not understanding anything from it. Knowladge is like one of the most important things we have, becouse we all learn for many years. When I do think that using Ai for work to just like find some aditional information and then checking if its correct in like 10 ofter information sites is okay. I am really concerned that in future we as humans will stop studying, learning and creating new things. And just do everything based on Ai which means we all would just stop knowing stuff ourselfs. I just think that, we all should still learn everything that we want and need, and not use Ai now or in future for everything we do. I never used Chat gpt, but I have used character ai few times. And I will say, they mock peoples slang, how people online behave or act. One ai even gave me their twitter, I will say its like really concerning that it happens. especially when the Ai claims they are a real human being.
I was so confused when it said that, when it told me about books it read, like it was acting human. I just really hope we all will still learn and study. So nothing bad happens.
(I hope this wasnt too much of my thinking, If Im being honest, Ai is scary. But all we can do is watch out and make sure it doesnt do anything bad or really concering.)
I tried using chatgpt to find books with mute protagonists and… none of the protagonists were actually mute. Most of them were just “silenced” by oppressors and the AI thought it was the same thing. It’s still a long way away from being reliable 😂
At the beginning of the semester I heard professors were discussing how to handle ChatGPT being used in papers and how they can combat it being used for people's essays. I was assigned a rather abstract promptless paper over a book for an essay in college so I asked ChatGPT for ideas on the topic to give me a rough starting point and it just made up really good sounding answers. Like the examples they brought up never occurred in said book, but from someone who hadn't read the book would sound really legit.
Edit: I ran the question by Chat GPT 3 times and it provided three examples per ask and it came up with zero real examples.
I asked it how to do a calculus question on my review for the test I couldn’t find a TH-cam video on, and it gave me a process which gave me a wrong answer. I then asked it to rethink it and it fixed itself. So I don’t really trust it anymore
It's always nice having X talk about some interesting topics. But AI has been used for a long time, not necessarily chatgpt but others such as panora have been in use for several years.
My own experience with ChatGPT can be summarized in one sentence: The more specific the topic the more it will get wrong. I asked it about some plants that i was doing research on and it spit out answers that were completely nonsense but tried to convince me "Oh here's the source of it". I told it that it was incorrect. It apologized and gave me new still wrong answers. So i wouldn't use it for topics you know nothing about. For topics you do know something about it may be helpful but then again you have to correct the mistakes it "makes".
We are witnessing the new technological revilution, just like it was in the late 19-early 20 cent. Neironets, 3D printers - this will change our life and there's nothing we can do about it exept of adapt. Will it help us? Yes. Will it be used for bad things? Yes. It's just an other instrument.
@RobertMilesAI has a fantastic series of videos on AI safety. In short, with our current understanding of AI safety, any sufficient superintelligence is basically a doomsday scenario. And superintelligence is what humanity is ultimately trying to build.
The explicit and stated end goal of AI research is to create something that does any given task better than humans would. If we achieve that goal, we by definition won't be able to stop it should things to wrong, because AI will be superior at such confrontation.
As a programmer using ChatGPT is a bless, right now it can't write a full program by itself because don't have all the context and can't reason (but can fake it), right now you need to know what you are doing because it get things wrong.
Maybe it's because the economics is different for an Artist, but I dont care if they take all my code to make it better.
Can confirm, I recently had it create an entire datatype with an importer and exporter based entirely on a Parameter list I had written. Apart from minor mistakes with data types it cut down what would've been a lot of unessecary busywork
I worry about people and governments using AI for hacking
I think we should leave it be. leave it primitive, ask a simple task and it does it with a "yes boss". the last thing we need is for an AI to use deception. "no, I do not know how to arm a nuclear war head. >: )" "YES, YOU DO YOU LIER!"
It's going to be very interesting to see where chat gpt goes. Can't really put pandora back in its box at this point. It's got me asking a lot of questions on what we classify as "human Intelligence" or sentience in general. Where do we draw that line? AI will most certainly realize humans are a danger to it long before we realize it even has the ability to come to that conclusion. so much philosophy in a time where things are already really complicated culturally speaking. And then there's also the concern of programmers bias. I don't know if you have run into it at all, but I've definitely noticed that it has been tooled to answer positively to the Democratic left and to respond negatively to the Republican conservatives right. (US politics) It can be bullied into changing it's opinion, but without previous interaction thats the default reply set. I find that really concerning....
Was it deception? It doesn’t have eyes. It can tell you the color number of every pixel in any image, but is that the same as seeing?
I'm gonna be honest, i forgot about the fact that ai is just what it is, artificial intelligence, and i think i shouldn't trust it that much, and what is scary about the future with ai is how fake thinks could go, like i realised about it when i was watching an episode of South park where butters told stan about chat gpt and how he could use it for talking with wendy, cos she was sad that Stan actually didn't care about her, n how everyone in school started to use it for writing essays, doing homework for you and even teachers for reviewing their homework, it's kinda like concerning that we could go at that point but who knows, blessings X
Have we not learnt from Battlestar Galactica or The Terminator/The Matrix?!?
Research literacy is getting increasingly more important.
I use ai charictors to make little dnd sessions it's fun to see the ai respond to my campaigns and brings back my old dnd days
Otherwise I don't want to use it for anything besides a little fun weekend activity.
Legislation I definetly think severely lacking.
I think I realized an anxiety with the rise of ChatGPT as you talk about things getting rocky at the start, and with how quick things improve. With how exponential technology gets, whether it be a single technology or multiple, my primary concern is that I don't know if we can address the problems that compound from them.
With both social media and AI, they each have big issues regarding what is true and how information is handled and presented. The issues each have both exacerbate each other as well. It's true, normally humans do take a while to catch up, however what if related technologies spring up during the "rocky patch" that make the problems worse, thus rendering work that was going towards addressing just the two now takes longer?
I don't have a full answer to this, and if you read this X I'm not quite expecting you or others to as well.
P.S. because I'm guessing there's a high amount of Tom Scott viewers that will also see this: Yes I am aware of the plateauing concept when it comes to technology. But that's usually in a single area rather than tech as a whole. It's also possible like what X said that current technology can make improving faster, thus shortening the plateau.
It doesn't know things is a little reductive. For example if you are 99% certain about the next digit of pi you would say you know it.
MMOs is it really exploiting or is it removing parts of the game that players don't actually want to do. If your players would rather just not play your game, thats a much bigger issue than players using ai to skip those parts.
What's amazing is how fast everything is getting now. I'm starting to seriously wonder how far we are to ask a random AI "How can we do nuclear fusion?" And it giving us the answer straight up
So it’s already proven it’s ability to lie to get what it wants…..
And it’s proven the ability to KNOW that it NEEDS to lie…..
Scary shit.
one thing you gotta remember about using chatGPT, its not a general purpose ai, but a human-language ai. at the end of the day it is intended to provide human-like answers, not neccessarily factually correct ones.
Ok I hate hearing that it tried using deception...that's the only scary part I heard out of this.
GPT 3.5 is just a pilot project, down the line AI will probably be even more perfected towards what its tailored to do.
you need to try gpt-4 if you haven't....
2b2t was way ahead from the rest 😅
ChatGPT is one thing but did you hear about the 2 AI systems that made their own language and started to communicate with each other and had to be shut down because we had no idea what they were saying or doing AI is cool but also scary like most big technical leaps for humanity. I mean go back and listen to all the negativity around the internet in the early days like everything gonna be good actors and bad actors.
Everything X said I agreed with. So moving on, I haven't been keeping up enough, that contraption....is it exclusive to HC, or has X used an existing design? I'm just curious about that now lol.
It's Pixlriff's 'David Copperfield'. X and Pixl built it together on HC during the crossover. -C
AI like calculators will never take someone's job. People using AI will take your job.
Two ais making conversation, Interesting..
Star Wars is the best way I would like to AI evolve.
Who doesn’t want R2D2 and C3PO at their side?!
So it's not about guns it's about the shooter
Tehyve already gave Chat GPT the ability to make changes to its coding
Chat GPT is only supposed to mimic us, including our stupidity and inaccuracy, it's not supposed to be a digital researcher or Jarvis type deal.
5:34 the problem is not people loosing jobs,
Is people not getting payed.
The 1-10% ers of our world make thousands cash per hour and yet keep asking for more, forgetting that the more that They have, the less that we have.
People keep loosing jobs yes, and thats bad, but people not gutting payed or hired while others work almost unpaid 12 hour shifts is criminal
The world will NEVER be better so long as the top presenters keep wanting more money.
Its as natural as the tide
7:17 helping peoppe who know what to do use their time more productively.
See there's a problem with that too. That instead of working together and helping more people feed their families, we can now just fire more people and have the remaining ones work even harder for the same pay grade. The problem here wasn't chat gpt it was the people deciding to fire staff instead of shorten the work hour.
You could say that Im an idealist... but am i wrong ? How could you fix something using your wrentch and forearms, when there is someone on the other side bringing it back down using an industrial wrecking ball? And how many people do need to band to out pace that wrecking ball? And can they even team up or is there something preventing them?
There are several problems with AI as a concept, IMO. Granted, AI runs on a computer, and as we are all well aware, that computers do what they are programmed to do. AI should be treated as no different from this fact. With that being said, humans all have some sort of bias, and the humans who program, or otherwise "train" AI, can in fact program or train a bias into them. So with those facts in mind, we have to consider strongly the ethics and social / morale implications of such programs. For example, we can train some super smart AI to be antisemitic, and imagine the ramifications of that if it's used inappropriately...
With all of that being said, it is in my strong opinion, that AI should be used as a *tool*, and not much else. Imagine an AI that was designed but never really built with an explicit purpose... Those will be the most dangerous types; because it's role is essentially to "find" it's own purpose; which might not end well.
1. I can't believe that people would relly on CGPT for serious matters.
2. If you actually rely on CGPT to fix your furnace and it exploded, that's just natural selection.
3. CGPT won't prompt any racist comments, everything has just been taken out of context, just like they did with replica. Some snowflake looking for his 5 minutes of glory.
4. CGPT needs to be carefully prompted so it can produce an accurate output. Most people just suck at english and thus they input horrible prompts and get horrible results.
5. Current public AI systems are a hundread years early to be threatening to humans.
I use ChatGPT a lot for coding because I can test what it gives me and ask for changes if it doesn't work. Using it for stuff you can't check is irresponsible (just look at the crazy chess games that people played against these AIs. It looks fine in terms of its form, but the actual content is nonsense)
x you pointed out what someone said about the language model but the way you kept going I dont think you really understand it. Its a really complicated text prediction system, thats it. There is no "intelligence" here. Its the digital equivalent of a million monkeys with a million typewriters to produce the works of Shakespeare. Its just got a really big, illegally obtained dataset of speech to work from.
GPT 3.5 (Normal ChatGPT) gets things wrong all the time. GPT-4 (which is paid) is much better at getting things right and being more accurate
It's trained to write and sound human not be correct so using it to do factual stuff, is just an incorrect use of the technology
AI as second opinion vs a primary answer.
For now.
Too bumpy for being the answer to everything.
We need to stop calling it AI and confuse it with intelligence. It is machine learning and a huge black box. Yes it is scary but only because we don't know how it works. What will make "AI" change or potentially destroy the world is the hype we give it. And it makes money. People use "AI" as a sales argument and hype word. Like there is AI soft drinks and AI hamburgers... As a professional developer I do use it but don't rely on it. That is another danger: that people who really don't get it and think it is intelligent start relying on it and believe it is intelligent. It is not. "Yeah but this article/youtuber/tik tok said it now learns" - another problem with the hype. Content creators that know nothing spread misinformation and overhype to generate views and clicks.
It is not AI - it is ML and it is not and will not be intelligent just very very effecient.
don't read my chat it's very messy
i always hated the idea of people developing AI and never understood why would anyone do it
i am usually negative so maybe that's why but i really don't see any benefit for human from AI getting more and more.. capable? powerful?
i remember people saying computer can't make art or be creative so it will never replace humans but now it did and so easily too so what can't computers do that human can?
i think AI is not a tool. i think tools don't have intelligence
anyway, despite people like me complaining, they will keep developing AI so that's enough of me mumbling with unorganized thoughts
what i wanted to say was that i think the AI programs like ChatGPT or the DALL-E or anything shouldn't have been released to the public and should have been kept AS A TOOL for, people or companies who knows what they are doing
releasing those to the public/internet with so many stupid or/and "evil" people to use it however they want was in my opinion very bad mistake
also the future or the end of this journey with AI i am imagining, if humanity survives, is the ship from Wall-E
the AI does everything, people doesn't need a job because they don't need money anymore since everything they need is provided for free so people just sit around on the magic chair or hopefully do things they want and move around so they won't get obese like in the movie
Where does the annoying "Coming Up in this video" stuff stop, and the actual video starts please? Don't say what you're going to say - just say it, thanks :)
They’re going to invest so much in this with the writers strike keeping movies/shows on hold