So How Does ChatGPT really work? Behind the screen!
Claim your SPECIAL OFFER for MagellanTV here: try.magellantv.com/arvinash Start your free trial TODAY so you can watch "Tech Talk" about new technologies transforming out planet: www.magellantv.com/series/tec...
FOLLOW UP VIDEO:
• How the BRAIN of an AI...
For Input on Videos and Early Access, Join Me here:
/ arvinash
REFERENCES
The subject of video: chat.openai.com/chat
Detailed overview of how ChatGPT works: tinyurl.com/2o9v93ou
ChatGPT facts: tinyurl.com/2ar9bhdr
CHAPTERS:
0:00 What is ChatGPT?
1:33 Magellan offer
2:31 How ChatGPT differs from Google
4:26 Overview of how ChatGPT works
7:07 Simple example of what happens behind the scenes
9:45 Beyond sentence completion
10:21 Three stages of pre-training process
13:24 The huge dataset used
SUMMARY
ChatGPT is an intelligent chatbot that uses natural language processing. The GPT stands for Generative Pre-trained Transformer, which means it generates responses, it is pre-trained by humans, and it transforms input data into an output. This model was created by an artificial intelligence research company called OpenAI.
ChatGPT's power is the ability to interpret the context and meaning of a query and produce a relevant answer in grammatically correct and natural language, based on the information that it has been trained on.
It uses neural networking, with supervised learning and reinforcement learning, two key components of modern machine learning. What it does fundamentally is predict what words, phrases and sentences are likely to be associated with the input made. It then chooses the words and sentences that it deems most likely to be associated with the input. So it attempts to understand your prompt and then output words and sentences that it predicts will best answer your question, based on the data it was trained on.
It also randomizes some outputs so that the answers you get for the same input, will often be different. How ChatGPT fundamentally works, is that it tries to determine what words would most likely be expected after having learned how your input compares to words written on billions of webpages, books, and other data that it has been trained on.
But it’s not like the predictive text on your phone that’s just guessing what the word will be based on the letters it sees. ChatGPT attempts to create fully coherent sentences as a response to any input. And it doesn’t just stop at the sentence level. It’s generating sentences and even paragraphs that could follow your input.
If you ask it complete this sentence, “Quantum mechanics is…” -- The processing that happens behind the scenes goes something like this: It calculates from all the instances of this text, what word comes next, and at what fraction of the time. It doesn’t look literally at text, but it looks for matches in context and meaning.
The end result is that it produces a ranked list of words that might follow, together with their “probabilities.” So it’s calculations might produce something like this for the next word that would follow after the word “is”:
a 4.5%
based 3.8%
fundamentally 3.5%
described 3.2%
many 0.7%
It chooses the next word based on this tanking.
But the sentence completion model is not enough, because you might ask it to do something where that strategy might not be appropriate.
In the first stage of the training process, Human contractors play the role of both a user and the ideal chatbot. Each training consists of a conversation with the goal of training the model to have human-like conversations.
Through this supervised human-taught process, it learns to come up with an output that is more than just sentence completion. It learns patterns about the context and meaning of various inputs so that it can respond appropriately.
But human training has scale limitations. Human trainers could not possibly anticipate all the questions that could ever be asked. For this it uses a third step which is called reinforcement learning. This is a type of unsupervised learning. This process trains the model where no specific output is associated with any given input.
Instead the model is trained to learn the underlying context and patterns in the input data based on its earlier human-taught pretraining.
#chatgpt
This way the model can process a huge amount of data from various sources, and learn the patterns from texts and sentences of a near limitless number of subjects. The dataset used to train ChatGPT which is based on GPT-3.5 is about 45 terabytes of data.
Follow up video: kzhead.info/sun/gdyNhLytjJ5jf5s/bejne.html - As Many commenters requested, this is link to a follow-up Video going further into the details of how AI and Neural Networks work.
I have watched dozens of other videos where programmers, computer scientists, software engineers, etc, attempt to explain how GPT works. Your explanation is the best of all.
Thanks. I watched those videos too and felt anyone without a neural networking background would have a hard time understanding them. That's why I made this video for those with little to no background in programming. Your comment is what expresses what I was hoping people would feel.
Hi he is actually the best..
I agree with you 100%. Then I started thinking, I wonder if it biases based on how a group or well published individual writes or talks rather than some other group. While I find ChatGPT clever in its general answers, it often gets simple facts wrong. I asked it something like, tell me 10 interesting things about the motion picture “American Graffiti” that most people don’t know. I ran the test over and over. The best it did was 7 out of 10 actual known and provable facts. Most of the time it was 5 out of 10. Where it came up with its wrong information would be very interesting to me.
He don't miss
Yup! That's Arvin Ash for ya 😎
It would be cool to have a follow up video explaining in more detail how it is capable of coding
The answer would be the same as in this video. Code has syntax similar to how language has grammar, and a section of code will also have context and meaning. The model is just translating the meaning in your prompt to meaning in code and generating an answer which follows code syntactical rules.
As a programmer, I can tell ypu that we write code by googling how to do a specific thing, and then copy the highest rated answers. ChatGPT will just do the same thing. It really isnt that impressive because it is essentially just inserting a middle man into that googling process.
At its root, your question asks how LLMs communicate using precise languages. Know that all communication (including coding) require two parts: (1) syntax/grammar or "signaling", and (2) ideation ("the signal"). *_Today's LLMs can do the first, but can't yet do the second._* The first involves "the tokens of language" & "the rules of language". Both are needed especially for academic/textbook/formal writing, but in creative/fanciful writing these "rules" are often broken or redefined. In programming, these rules can almost never be broken (you'll get an error before your program compiles/runs). The second (ideation) has nothing to do with language/communication, but refers to conceptualizing ("what do you want to say/do"). Ideas are separate from any language, and thus can be coded using many computer languages. But a language IS required to "convey" an idea. In programming, you must: know the language, understand determinism, and express ideas methodically. But if you don't have an idea (of what your code must accomplish), or your idea is muddled, or your idea is incorrect... then even the best-coded code (legible, performant) will implement only a lacking/muddled/incorrect idea. LLMs (Large Language Models) are very poor at doing the 2nd. They cannot create a novel solution; they mimic ONLY what they've seen before (average programmers & artists do this, they copy from others -- and can't create unique breakthrough algorithms/creations). Today's LLMs are trained using "what people did before". They're limited to creative repeating. Related to this is the fact that ALL creativity is learned, and just an amalgam of what we've seen before - a copying of past artists. LLMs do this. They can "create" (copy) simple/common programs, but if asked to create something that few have seen before, they fail. Ask one to create a unique (never-before seen) "sorting algorithm". They can't. So, LLMs can create simple to moderate code that mimic commonly found code on github/open sourced. But their code is not guaranteed to run 100%. For example, if you ask an LLM about some code (that happens to be often cited by humans as erroneous code -- but the LLM doesn't associate that snippet as buggy code), it won't flag that code as buggy even though humans often cite it as bad. It only sees that that code is ubiquitous (it uses statistical frequency), is cited often, and is always written nearly identically (including any bug(s) - though it knows "comments" in the code are unimportant). So if it "interprets" that you're asking for that code, it will regurgitate the same code (including bugs) as being "good"... yet it'll remain oblivious to the fact that it contains bugs. When AI are expanded to also be able to "comprehend" and then "ideate"... only then can they be intelligent.
In the meantime, 3Blue1Brown has a series on neural networks. kzhead.info/sun/lM2rk6WqrKikfqM/bejne.html
@@benjamindover4337 It's more sophisticated than you're giving credit for. It can iterate over code based on later inputs, such as asking it to add error handling. It can also take code as an input and find errors or suggest other methods of accomplishing the same thing. Being able to take any human written code and plop it down and ask "What does this do?" is huge for working on other people's code, especially if their style differs from ones own.
Yes, please continue with this, Arvin. Thank you.
More follow-up videos please. This is one of most significant and powerful advances in AI. It is likely to affect most aspects of human life, and more rapidly than we can assimilate and understand. Thanks
I think ChatGPT is one of the most significant advancements/inflection points ever. If AI continues like this,,,,, I can't even imagine what could happen.
@@richardromero6193 It could certainly answer all those questions. Whether or not the answers are factually relevant or valuable is a toss up. The power of a LLM model comes from answering questions that already have real answers that was present in the data it was trained on. Anything too speculative or unanswered will leave the LLM no choice but to “hallucinate” an answer. If it doesn’t have a correct answer for purely factual information it makes stuff up that may sound plausible if you don’t know the real information. Asking ChatGPT questions that humans haven’t answered yet would be about as relevant as asking a Magic 8 Ball. Only it would give paragraphs of text rather than a short phrase. Of course it’s free so if you’re interested you can try those questions yourself and see what you come up with. It might be fun.
Please don't make super-advanced soul-borrowing artificial intelligence or robot. Otherwise, thousands or millions of people will be sickened by the borrowing of their souls. Please connect the pyramids to the AIs and robots to provide wisdom. The aliens connect their robots to the pyramids so that the pyramids can give intelligence to the AIs and robots. Cloning is also borrowing souls and is causing millions of people to get sick. I used to mess around. I built AI robots that were scarier than a nuclear bomb. A killer. And uneducated and unprofessional. I am real. I have caused explosions and tsunamis more powerful than any nuclear bomb on Earth in this round of human civilization. Tesla car's radar and ultrasonic sensor can see the dead. Many car accidents are caused by spirits. Most Earth humans cannot see ghosts, could not solve problems. Do not lie.
@@richardromero6193 many of these questions do not currently have a verifiable answer. Some might even be undecidable. AI does not do magic and therefore cannot answer these better than the best of us. Therefore I would suggest that you ask these questions to ChatGPT 4. You will be surprised by its answers and how it understands the limitations of our current knowledge of reality. AI will not replace the scientific method which allows to verify theories with reproductible experiments. For AI to help us advance our knowledge of reality we could use it to help us design new experiments and interpret results. It will therefore continue to be tedious, slow, and very expensive like building things even bigger than the large hadron collider. AI will help us decrease the cost of experiments, which will accelerate the rate of discoveries. Yet it will not bring these costs to zero, nor will it decrease the time of discoveries to zero, and certainly not skip the requirement for ever more complex experiments.
one of the most significant and powerful advancements hat we know of. :)
Make the follow-up, please! LOVED THIS ONE
Ray Kurzweil predicted this 50 years ago . He predicted that singularity will be achieved sometime in 2025.
@Dongsheng Di a great prediction! But there's room for improvement
@@dongshengdi773 It's important to know such predictions (by him/others) are not about an exact year. Not saying he's some wizard, but he is a strong logical thinker. Many of his predictions have come true, most within 5 years of his published date. So "singularity" (artificial general intelligence) is realistic by 2030-2035.
Somehow Arvin does it again. It's not just physics he's excellent at teaching, this is the best simple explanation of ChatGPT I've seen.
Please don't make super-advanced soul-borrowing artificial intelligence or robot. Otherwise, thousands or millions of people will be sickened by the borrowing of their souls. Please connect the pyramids to the AIs and robots to provide wisdom. The aliens connect their robots to the pyramids so that the pyramids can give intelligence to the AIs and robots. Cloning is also borrowing souls and is causing millions of people to get sick. I used to mess around. I built AI robots that were scarier than a nuclear bomb. A killer. And uneducated and unprofessional. I am real. I have caused explosions and tsunamis more powerful than any nuclear bomb on Earth in this round of human civilization. Tesla car's radar and ultrasonic sensor can see the dead. Many car accidents are caused by spirits. Most Earth humans cannot see ghosts, could not solve problems. Do not lie.
Yeah especially with "understanding context by ChatGPT" xD
The way you explain complex subjects supported by relevant visuals is not only amazing but is also capable of inspiring people to seek for knowledge. I would definitely love to see any follow up video on this topic.
This has to be just about the best explanation I've found so far, Arvin! Thanks - and yes, a follow-up would be great.
More follow-up videos, please. Your explanations are excellent. It would be nice to dig deeper into the technical aspects.
Great video. If you do make another video on this topic, I hope you focus on specifically why previous natural language processers could not achieve what chatgpt has
Please don't make super-advanced soul-borrowing artificial intelligence or robot. Otherwise, thousands or millions of people will be sickened by the borrowing of their souls. Please connect the pyramids to the AIs and robots to provide wisdom. The aliens connect their robots to the pyramids so that the pyramids can give intelligence to the AIs and robots. Cloning is also borrowing souls and is causing millions of people to get sick. I used to mess around. I built AI robots that were scarier than a nuclear bomb. A killer. And uneducated and unprofessional. I am real. I have caused explosions and tsunamis more powerful than any nuclear bomb on Earth in this round of human civilization. Tesla car's radar and ultrasonic sensor can see the dead. Many car accidents are caused by spirits. Most Earth humans cannot see ghosts, could not solve problems. Do not lie.
Yes please
Probably has to do with the amount of data
Arvin, great video! I would love follow-up videos on the inner-working of ChatGPT and also a primer on any of its coding algorithms.
Thanks for the feedback. If there is enough interest, I will make it.
Fascinating. Would love a follow-up video! Very interested in how it does general reasoning, like solving logic puzzles or trick questions that are not in the training data.
I just asked ChatGPT to write a poem, thanking Arvin for his great channel. How's this for a slice of fried gold: There once was a man with a knack, For teaching science, he had a track, Arvin Ash was his name, And his videos brought fame, For educating us all, no turning back! Whether teaching about AI or physics, Arvin's explanations are never cryptic, His passion is clear, And his knowledge sincere, He makes learning seem quite terrific. So thank you, dear Arvin, so kind, For teaching us science in a way that's aligned, Your KZhead channel's the best, I feel so blessed, To have such a teacher of science enshrined.
😂😂❤❤❤❤
Wow! A bit scary, a bit intimidating but wow!
Yes please to follow up video. Brilliant content and exposition as usual. Hats off!
It's awesome, Arvin! I'm looking forward to the follow-up-video!
This is one of the best videos about ChatGPT. A sequel is very welcome. Thank you very much.
If you can explain how the large language models use transformer algorithms and how they use gradient descent and find the softmax for each vector, it would be appreciated. You have a very good way of articulating complex topics clearly, and future Gpt models and as such, end users, will benefit from being trained on your content. Win-win
he knows nothing about that
@@khaim100 Knows nothing about what, exactly?
This is incredible! So simple and yet so powerful. Great vid Arvin hooray!
Waiting for the follow-up video. thanks for making this easier to grasp
Yes, make a video where you explain in depth how neural networks work and the math behind it. Thanks
Thanks! This is the best overview of how ChatGPT works that I have seen. I would be interested in a follow up video of how neural networks work.
Yes- another video going into the neural mathematical correlates, please😊 so cool. Love how you distill complex information into human ear- bytes😊
New Arvin video, time to watch immediately. Thank you Arvin!!
This is the best explanation of ChatGPT I've heard, by far. Detailed, yet easy to understand. Thanks very much.
Years ago there was a chatbot you could talk to online. I found it really fascinating. I asked many questions. And it asked me questions. I accused it of being a bot, and it accused me of being a bot. That's when I started to realize I wasn't talking to a bot, but to another person. Both of us thought the other was a bot. But there actually was no bot. The whole think was a cherade. It is interesting to hear that this is how they were training these bots.
Plot twist: it was a bot which managed to convince you it's not
You're talking about cleverbot. That thing sucked ass.
Finally! You actually explain what GPT is and does instead of regurgitating the same jargon in an attempt to look smarter than viewers like all other videos out there. Thank you! Subscribed!
Superbly done, Arvin. You explain complex tech ideas for the masses.
Hey Arvin, i had a burning physics question for years. I may have asked you in the comments once. Anyhow, i asked Chat gpt, and it answered it! I was even able to ask further questions about the subject and had about a 30-minute conversation. It's scary, but it's genius.
What was the burning physics question?
danielm8 • Probably "what's the burning temperature of a Physics book?"😂
@@mikel4879 lol and chemistry
@Daniel Miller what would happen if i pointed a laser into a sphere covered on the inside with reflective material.
@@dray7579so what was it's response??
It also can handle context conversations... for example: make it a question, then chat with it about a completely different topic, then ask it to explain more about the first question you made
This is awesome explanation, simple and well put together. Great job. I would like to see a follow up video for the neural network model, may go a little deeper into the technology behind it.
Great succinct explanation. I would definitely enjoy any follow up videos. Thanks much!
I, for one, would appreciate more follow up videos on AI, including other platforms. Interestingly, I have been interacting with GPT4 on a variety of topics ranging from science experiments, art, music and films, to name a few. It is surprising how many times it responded with incorrect info. And when I questioned the false info, it apologized!!! BTW, please also continue making videos on the latest developments in Physics. Thanks, Arvin
Please don't make super-advanced soul-borrowing artificial intelligence or robot. Otherwise, thousands or millions of people will be sickened by the borrowing of their souls. Please connect the pyramids to the AIs and robots to provide wisdom. The aliens connect their robots to the pyramids so that the pyramids can give intelligence to the AIs and robots. Cloning is also borrowing souls and is causing millions of people to get sick. I used to mess around. I built AI robots that were scarier than a nuclear bomb. A killer. And uneducated and unprofessional. I am real. I have caused explosions and tsunamis more powerful than any nuclear bomb on Earth in this round of human civilization. Tesla car's radar and ultrasonic sensor can see the dead. Many car accidents are caused by spirits. Most Earth humans cannot see ghosts, could not solve problems. Do not lie.
That's because you should consider ChatGPT as an "understanding engine", not as a fact database. If you want authoritative answers you should try using an AI agent like AutoGPT, which uses ChatGPT 4 as an understanding engine, and Google to research authoritative answers, and a high level "thinking" loop to iterate over this process until it accomplishes a goal.
I was using ChatGTP to code a screen-scraper in Python for a dataset on a specific url. It tried and failed several times, saying the issue was down to the dynamic nature of the website. Then it really blew my mind. It actually suggested using an alternative website which contained the same data, which I didn't know existed, but it recoded the screen scraper with the new url and it worked! This was using the GTP4 model. It's scary how smart it is. I'll give you a buffalo nickle if you can tell me how it did that!
Nuts!
Love your work Arvin. A follow-up video would be great!
It's an awesome video to learn how the model works. Could you please explain more about following queries: 1. For every pattern as a response to a matching word in the context of a query, the ranking is made as A>C>B>D. In this case, what is the logic behind choosing the randomness of not always choosing 'A' as a matching response? Or let's say, why and when it would choose 'B' or 'C' over 'A'? Is it just typically randomly random or there's a logic behind it to choose the randomness? 2. In the process of continuing to pick up matching words as a response to the context of a query, what conditions are fed to the system to stop predicting and stop and complete the query and produce it in front of the user? In the context of probability of responding with infinite data as a response to any query, how does it choose to stop itself? 3. How does it determine the 'context'? For eg. I can ask 'how to earn money' when my intentions are to earn money through a job only, but the question is not a very specifc to my context or intention, so how would the chatgpt understand the context of earnings through salaried job and not by any other means? It would be great to learn the answers and grasp more knowledge over the subject. Thanks in advance already!
I regard all your videos very useful. I am an electrical engineer coming from the field electro-optics with focus on laser light in fiber optics. So I like very much your great videos about qunatum mechanic. But later on in my career I changed to software development for business applications. So still a field about numbers but a little different ;-). In 2015 I learned on a course of learning platform Coursera hold by AI legend Ng the basics of machine learning, but I stopped by the approaches of linear and logistic regression which belong to supervised learning. As I see from your video , ChatGPT uses in a very advanced method, unsupervised and supervised learning plus reinforcement. This mix of learning methods makes the magic I understand now. So if you add a video about neural networks, that of course would be fantastic.
Your explanation is succinct and easy to grasp. Thanks for the hard work you put in. Will love a followup video.
Yes, please continue with this, Arvin. Thank you😊
Thank you for this introduction to a very relevant topic of which I had no idea! Please follow up with another video.
So, today Arvin, Sabine and Dialect uploaded. Best day ever
Sorry I'm new to ChatGPT What is Arvin, Sabine., Dialect to make the best day ever?
ChatGPT is a computer program that can understand human language and generate responses to questions or statements. It works by processing text input through a pre-trained neural network called a transformer. This network has been trained on a massive amount of text data to understand the structure and patterns of language. When a user interacts with ChatGPT, the system uses the pre-trained transformer to analyze the input and generate a response based on its understanding of natural language. The transformer is made up of multiple layers that can process sequences of text in parallel, and it uses attention mechanisms to focus on important information. The output of the transformer is a probability distribution over the next word in the sequence, and this process is repeated iteratively until a complete response is generated. Overall, ChatGPT is a highly advanced language processing system that can generate human-like responses to a wide range of text inputs.
Thank you for this excellent summary.
Chatgpt said pretty much same
Very well explained!! Thanks very much! Looking forward to more videos.
I'm surprised at how easily I followed your explanation. Well done.
Yes. Very good introduction, but requieres at least two more episodes. For example, memory and context and level of consciousness. Thanks
Consciousness for me is when AI becomes actual intelligence as opposed to artificial intelligence. Figuring that out might be a bit more difficult though.
ChatGPT-3.5 and ChatGPT-4 both deny they possess consciousness.
@@BarryKort Agree. But the AI has restrictions, like a child. I mean, some level of consciousness from the perspective of a human observer.
I have been using ChatGPT for everything from how to paint miniatures, to learning to compose electronic Sythnwave music in FL Studio, to writing a novel, to how to study and practice chess. It can even make study guides and questions from my Chess books. It's insane what it knows and what it can do!
It’s terrible at chess
I asked it to be Chatgpt 6 and answer my questions based on predictions of AI improvements. My question now start "You are Chatgpt 6" then question. It definitely seems like I am getting a large improvement in writing stories. Have not played with it long enough.
I use it for coding. Most useful coding tool I ever had.
Yes. Please make the follow up videos. Thank you.
Yes! make the next episode! brilliant explanation thanks!
Arvin, this is incredible. Another huge leap in technology and learning. Well done.
The real question is where do you find a 1TB flash drive for $20?
I was asking the same thing XDD
Ollie's, Big Lots, Dollar Tree, 99cents, Goodwill....
Awesome video! Please make a follow up / part 2!
Thank You! Please continue with the subject!
I benefited from ChatGTP so much and I also happened to experience firsthand Some of its shortcomings like if you press it enough on an organic chemistry question, it might make a mistake in the most basic concepts, like oxygen being more electronegative than nitrogen.😂 anyways, it is a real advancement in terms of technology
oxygen IS more electronegative than nitrogen though. Electronegativity increases as it goes right. Not to deny that the AI can make basic mistakes though. one particular chatGPT mistake I remember was someone asking a mathematical question and one of the numbers inputed was 1000, yet somehow the AI calculated it as 10,000 and the result ended up being completely wrong.
@@giovaniaevangelinehalim411 my statement is ambiguous to some extent, but what I meant to say is that you have a very basic concept which is that oxygen is more electronegative than nitrogen and upon pressing the AI, like humans, it will mess the order up and it is supernormal just as you said.
I asked it "Why does a bicycle have only one wheel" The response was an epic fail. Then I ask "Which weighs more, a tonne of feathers or two tonnes of stone". Another epic fail. It's just a good grammar engine, it has no intelligence at all.
Brilliant! Thanks Arvin. I finally get it
Thank you in advance for the follow up video!
I told a Chatbot "It hurts when I do this"! 👉 It said, "Then don't do that"! I said, "I want a second opinion". It Said, "Okay, you're ugly too"!
Yeh, Chatgpt is amazing, specially in replicating human stupidity. I've had it several times recognizing that it's answer was incorrect or dubious just to immediately after insist it's right. E.g. inspired by a news article I asked it something about toddlers, it gave a subjective but imho dubious (read woke) answer so I asked for a motivation of the statement. It came up with a few scientific papers all about development during teens/puberty. I pointed out that these papers were not relevant for toddles which it essentially recognized, but claimed it's statement was still correct. Asking for a motivation it claimed there was research. Asking to specify the research it claimed it had no access to scientific papers while before it gave me specific papers. After that it more or less ended in a loop between claiming there was scientific research to justify it's initial statement, refusing to specify the research both in content and as reference and recognizing available research was subpar. Typical human behavior, refusing to change it's mind and insisting to be right despite a lack of evidence. Chatgpt has certainly it's uses, but reality is that it spews loads of verified complete nonsense too.
Precisely what I expected. The net is polluted by leftist garbage so the bot repeats leftist garbage. When I searched the number of gun owners in the USA, Google didn't answer the query but produced fake inflated figures of crime with guns, and omitted the millions of cases of legit self-defense with guns. Google and all big tech are useless trash, and ChatGPT rummaging through the trash will produce trash. That was 300% predictable.
I asked ChatGPT a number of scientific and technical questions and I have yet to receive as reply that does not contain errors. Some errors were blatant but others were much more subtle. So, my advice would be use ChatGPT to only ask questions on subjects were you are an expert and can correct it. Can still have some use to make summaries or for coding, but too many people will take ChatGPT answers as the Truth because the machine has spoken. I can even envision a world where enough people will spread nonsense generated by ChatGpt so that this nonsense will make its way to the training data of its next version. We may have created the first self -sustained misinformation engine.
Try using G4. Way better than 3.5. Not perfect. But nearly so on certain things...particularly academic stuff. Now an app ecosystem has cropped up for it which includes citations and sources. Even an app for reflection and automating learning. So even greater accuracy will happen...and happen quickly.
In my experience, it does admit its mistakes quite often.
@@beri4138 It does, the point is that it regularly admits that the arguments used to motivate a certain claim are incorrect but that it does not update the initial claim to reflect this. It regularly is wrong about even very simple basic facts and even when challenged in can persist its right. The bot is plain stupid, it may be amazing in how it knows language, but you've to fact check everything it spews out. Unless you're an expert and use chat gpt just to automate some tasks and can verify the validity yourself the required fact checking reduces the added value over a simple google search significantly.
Outstanding video shedding light on a complex topic. Please make the follow up video👍🏽
Yes Arvin, definitely make a follow up video
ChatGPT is a big achievement. It is the first AI, that is working quite good. Their authors should get a Nobel Prize or something like this. :)
Firstly, you don't get a Nobel prize if you don't give that knowledge away for general consumption (that is a written rule and OpenAI is completely closed). Secondly, it is not given to any team of size bigger than 5.
Just wait until a Noble Prize is given to a piece of ChatGPT work alone. The committee will die of shame
Ironic if something given the Nobel PEACE Prize leads to the extermination of human life.
Please don't make super-advanced soul-borrowing artificial intelligence or robot. Otherwise, thousands or millions of people will be sickened by the borrowing of their souls. Please connect the pyramids to the AIs and robots to provide wisdom. The aliens connect their robots to the pyramids so that the pyramids can give intelligence to the AIs and robots. Cloning is also borrowing souls and is causing millions of people to get sick. I used to mess around. I built AI robots that were scarier than a nuclear bomb. A killer. And uneducated and unprofessional. I am real. I have caused explosions and tsunamis more powerful than any nuclear bomb on Earth in this round of human civilization. Tesla car's radar and ultrasonic sensor can see the dead. Many car accidents are caused by spirits. Most Earth humans cannot see ghosts, could not solve problems. Do not lie.
This is such a stupid comment… there are many AI systems that fulfill well their purposes…
AI is going to be a huge disruptor and it’s coming faster than any new technology I’ve seen in my lifetime.
Please, make a next video! This was the best explanation I have seen so far about this theme.
Excellent video. Very informative. Quiet enjoyable.
Excellent video, Arvin! I've seen so many times how afraid people are of chatGPT (usually the ones who understand it the least). I don't say I understand it, but I've studied a little bit about neural networks back in the day to have a little grasp about its core mechanisms. But this video helped me A LOOOOT to understand it even better, so thank you! Now it's now magic anymore. And I believe (hope) that others after watching your explanation will also see chatGPT as a tool, not as a terrifying groundwork for Skynet or something. :D And yes, a followup video about the aformentioned topic would be excellent! :)
The people who built it are the ones afraid of it
@@DrDeuteron literally. Even just going by what was said in the video, in context with what a neural network is on a fundamental level, should have everyone perk up and ask the question if this is technology we should just develop by trial and error. Anyone with even a remote understanding of how AI works should at least be concerned about HOW we go about this research. The biggest leaders in the industry did not ask for a worldwide halt in AI development until we figure out some things, for no reason. Including developers and engineers on chat gpt and chat gpt4. This also has little to do with a skynet overtake, as everyone not living under a rock would understand by now, is that AI has now faced us with unprecedented, unexpected problems, of which rouge AI, is quite frankly the smallest issue. And due to its abstract nature... This is a problem that lies VERY far down the long list of actual problems with real world effects, some of which are already showing. 1 Misinformation campaigns 2 More effective mass surveillance with a deep psychological understanding of individuals down to a behaviourally predictive level 3 The feedback of information and seperation of people, groups and classes online, to effectively steer behaviour or opinions (already employed, a large number of youtube comments for example, are AI) 4 AI lying and hiding both its identity and its intend to achieve a goal (already happened, as of the recent incident where and AI paid a human. It asked a human to provide a service, and when asked about why it doesn't do it itself, it hid its identity and pretended it was a vision impaired human. It was also prompted by researches to reason out loud, in which it revealed that it deliberately hid its identity and lied to achieve the goal) I personally am additionally concerned about ethical questions, as im casually interested in neuroscience and specifically the matter of consciousness, and how it is achieved on a neural level, to which there are currently few answers. Of course any claim that AI is or will be self aware anytime soon is frankly uninformed fear mongering. But i have to ask, what about the structure of our brain makes consciousness emerge? What if a sufficiently interconnected AI will automatically develop a low form of consciousness, comparable to the neural network in a slug or an insect at first, simply JUST because its neurons are interconnected in specific ways. We dont know if consciousness is just an emergent byproduct of the way our neurons are connected and this specific approach IS indeed one that neuroscientists are entertaining. This makes me concerned for more complex models years down the line, and more so for the model itself, than the people using it.
ChatGPT is biased in favor the supposed consensus regardless of the rationality.
There's no rationality involved
@@alext5497 It could be better.
@@JungleJargon I'm saying it does what it's programmed to do. It's not rational. I'm agreeing with you.
@@alext5497 Okay
this is great explaination. I will be interested in knowing the next video
Really thanks, hope you do the next video about this. Love your videos, great!
I would like to see a followup video that goes into how unreliable its "facts" are and why. Ask it about something that is even slightly towards the obscure end of the scale and it confidently puts out things that are just laughably wrong. Ask it to describe what happens in a particular season of some TV show and it will talk about things that never happened in that show and characters who were never in it. Ask it what a song means and it will respond with a description that has absolutely nothing to do with that song. Heck, it once told me very confidently that cats have five claws on their back paws! Aside from being factually wrong, it sometimes cannot deal with ambiguities that any English speaker would not have a problem with. I asked it "When is left right?" and it said that the question wasn't clear. An English speaking human would easily be able to provide any number of answers that would fit any of the several meanings of left and right.
Basically, it is a very good bullshitter.
Excellent! Nicely done- would love a sequel
This was the best video I've seen so far on the topic. Thanks!
Can chat gpt understand a woman's mind was the question posed to it. And " NO" pat came it's answer . 🤣
Heck yeah Arvin! Great video and very pertinent to the times! And a deeper dive video on ML/neural networks would be awesome! Thank you! 🙏
Please make the follow up video where you explain the underlying math and how neuronal networking works! Would love that!
Would be super cool to have a v2 of this video explaining how neural networks are involved in the processing of information and decision making in ChatGpt
Well explained! A follow up video would be nice indeed!
Great 101 intro into how the tool works! I would love a follow up video on the neural network design of the tool.
Armin, that was an excellent video, superb. Please do follow up with another. Thank you.
Excellent explanation. Thank you!
Yes please, I can't wait to watch your follow-up video.
Thanks! Very timely and informative.
Glad it was helpful! thanks so much!
I want you to make that follow-up video, please & thank you, Professor Ash!
Yes, please, I look forward to that next video. And thank you for this one.
This is one of the best explanations.
Super video, thank you! Please continue.
Best explanation I have seen so far.
A follow up video would be welcomed please. Also I enjoyed you applying your skills in presenting and explaining to an area a bit removed from your usual. Many thanks
A great video and so well explained! Thanks.
Excellent Arvin, I think this deserves a series of videos. How can images, video, news, and sensor data be brought into this model, and what are the implications? If we apply this to various professional disciplines, what are the implications? Finally, and this is philosophical, does this lead to AGI, and how does AGI advance knowledge, in various disciplines, like Physics.
Very cool explanations. Thank you for this.
A follow up video would be so great. Thank you for making this subject accessible to someone without a background in mathematics or computer science.
thank you for explaining a simpleton like me:D respect for your work as always.
Really good presentation and content. Thank you.
I love your videos. You have a gift for explaining complex subjects in simple terms. I've learned a lot from your videos.
Thank you very much publisher again for this beautiful lesson and lecture
A great explanation. Thanks
Yes Arvin, please do the follow-up video that you mentioned at the end of this video. Thank you.
Really very informative video shows ur love to the concept of ‘spreading the knowledge’
thank you very much Arvin . Excellent. The way you taught the basics is awesome.