What GPT-4 Can Really Do
Вставка
- Опубліковано 15 бер 2023
- GPT- 4 is out and here are 2 ways it’s so much crazier than the last version that we all freaked out about. GPT4 is the model under the AI chatbots you can talk to - you might be familiar with ChatGPT or similar models like Bing Chat.
It can now ACE THE BAR EXAM. The version before this couldn’t do that. The biggest difference though is it can now analyze images...
If you're as blown away by this tech as I am, subscribe for more!
#shorts #gpt4 #ai #technology
Showed it a picture of my fridge and asked what I can make. It's only response was "Fried eggs"
It's not wrong.
You can afford eggs??? King
Is GPT 4 already available?
did you do it via the api? in the chat gpt plus ui it isn't possible to give an image yet.
It asked me if I was alright.
@@ricardomadleno564 You have to pay $30 a month for it, but yeah.
This is 13 years before the neural revolt of 2036
2036 is just 13yrs away? Wtf i could sworn it was 20yrs away
@@sathyanarayanan4839 2050 is as far away from now as the time between now and 1996
@@Adomas_B I remember reading books that said, "In th distant future of 2050." Now, I think, you mean in the next few presidential terms? There are members of Congress who have lasted longer than that time span. 😂
us 90's kid know that year 2000 is THE future!
you know.. flying cars, humanoid robots, cybernetic organ transplant, flying cars, flying cars
@@Adomas_B Almost one quarter of the 21. century is already gone.
“Chat GPT, change your name to Jarvis and get my suit ready”
I mean, if it’s integrated with an advanced 3-d printer and the right materials…that’s no so far fetched.
Change it to Skynet
i argued with it about this a while ago, we've come to an agreement that gpt IS jarvis
Lmao there's literally one ai on the app store named jarvis
the tech for all this exists.@@TWHowl
It’s really nice of Natalie Portman to teach us this stuff!
oh my god just realised she looks just like her
i don’t see it
😂 fr
Natalie Portman? I thought she was Keira Knightley 🤷♂
Portman can definitely play her in a movie.
I need my room to be completely cleaned. ChatGTP can’t pick up a broom
Soon enough bro, soon enough.
Terminator will come clean it for u
@@yengi0hhh if he can be reprogrammed to clean instead of kill and destroy THEN YES PLEASE
@@dmulugeta yes that’s what I meant by cleaning. To kill, destroy or even blow it up 🤣
Wait till roomba incorporates it
Looking at the tests Chat GPT took I noticed that it still failed the AP English exams, so point one for humans still
It is one point not point one
The reason it does so poorly is bc its trained through the interent, that exam is one of the only ones that does not require you to know information outside of what's on the exam. Critical thinking. Also Cleo failed to mention how gpt has been completely making up search results and information when being used as a search engine
@@ornettebreaker “there are big limitations here still” pretty sure that’s encompassing of the problems it still has but I must have misheard. Or misread.
Also failed the AMC 10. The baseline score you'd get for leaving the test blank is a 32.5 yet it managed to score a 30.
Gpt4 should have steered clear of urban dictionary.
Now Duolingo will actually know where you are
That one friend who's always excited about talking everything
FYI: The comments here were all generated by GPT-4... including this one.
As an AI language mo-
- ther fu-
-cker.
Moth er f*****
Noun
An insect whose kink is procreating in hospitals for moths
The ability of GPT-4 to synthesize the sum total of the data on the internet is amazing. I asked it to just speculate on potential transcranial therapies for epilepsy to replace intracranial electrical neural stimulation therapies.
It came up with a number of the same ideas that I did as a multidisciplinary engineer after talking about it for two hours with a neurologist specializing in epilepsy interventions.
It's going to find multidisciplinary solutions to things that people just don't realize. It can grab two disparate concepts from two different fields and understand how they might glue together. In seconds.
How on earth did you access it?
@@anuragC819
microsoft edge.
but it is a paid service to use it at its maximum capacity.
(chatgpt 4 is paid not the other versions.)
(you can still use chatgpt 4 on microsoft edge but it is not at full capacity.)
@@anuragC819 Microsoft Bing already runs on GPT-4. Just sign up to the waitlist, I only had to wait for like a week to get access.
@@anuragC819 you can sign up for open beta iirc
Uh isn't that what we call creativity???? Or is that not creative because it relies on us to ask the question?
I think humans will completely stop using their brain
Humans will stop existing in the first place
You say that as if that hasn’t already happened 😂
Just like that one doraemon
sorry I'll leave
I agree. Most people don’t see that work is actually a BLESSING. It keeps our brains SHARP. Not completely every work but more than half of all jobs need some kind of thinking. It’s really sad what this will do to us…
Too late.
Can it file my taxes?
I could prolly ace the bar exam as well. I'd just need all the answers and for someone to take it for me
I mean the fact that it even recognises something as a question and can produce an answer autonomously is pretty impressive
It didn’t even ace it. It got a 74.5%. It actually only scored 22 points over the minimum passing score.
@@alightswitch777 it got 90th percentile tho
@@alightswitch777 It got in the top 10th percentile of humans.
@@NosManJr They did not include any problems that were present in the training data.
My career is over before it even started ...
Edit: Mom I am famous
What do you study
A lot of careers will be over. Better start doing what the AI cant right now.
@@hermitcard4494 true but it's too late for me
@@hermitcard4494 It can do everything, albeit poorly, all by virtue of being a very fancy statistical model, not proper mind. Write code? Easy, just don't expect it to work for anything actually complex. Art? Yep, just get ready for some eldrich monstrosities for hands. Writing? Of course, just don't expect any original ideas, just some regurgitated concepts from the training set respoken in a monotone writing style. It's not enough to replace anyone doing the actual work, but good enough to allow your boss to threaten to replace your team with one of these things.
@@rkvkydqf and do you honestly believe technology stays static? People who used your argument many decades ago would have NEVER imagine the technology we have now.
could you go through the research paper done by Microsoft with the 15 points on gpt4 ? they conducted research on its capabilities before it was available to anyone else and with the unrestricted version
Gpt "prompt engineer" is already a high-paying job.
Edit: since this was posted my brother has gotten a job as a prompt engineer!
EDIT2: I HAVE NO IDEA HOW YOU CAN GET A JOB LIKE THIS, PLEASE DON'T ASK.
And it's really not that hard if you know some basic programming. It's basically writing pseudo code.
Which means it won't stay a high-paying job for all that long, because people will quickly get the gist of it, increasing the competition.
@@lonestarr1490 but you still need to think like a programmer, and most people think the internet runs on magic, so I'm not too worried about it lol
@@lonestarr1490 Yes. The thinking that prompt engineer was a new career was, in my opinion, wishful thinking, click bait material, and scam speak for "give me your money and I will commune with the almighty GPT on your behalf" . The reality will be closer to just another skill people already in house an any given busines will adapt and learn for their position. Anyone old enough to remember when stuff like excel came out will see what I'm saying.
Is it? I didn't know there was a "prompt engineering" job. LOL
Prompt engineering jobs won't be around long. You'll be able to say "help me out," take a picture of what you're doing, and it'll know from context how to help you. GPT-4 is pretty much there already.
We are rapidly approaching the point of singularity and it’s both fascinating and terrifying at the same time.
It's a language model it has no concept of self calm your tit
I agree.
The transcendental object at the end of time
What does that mean
@@gaaraofthedesert6566 the singularity is when technology develops so rapidly that there is no stopping it or return thanks in large part to AI. The Transcendental Object at the End of Time is a film narrated by Terrence McKenna about the end of human civilization. We are either A. Going to put laws in place and stop AI and tech at a finite point. B. Integrate or be eliminated by a superior AI race C. Create AI robots that will make humans never have to work again and all of the worlds problems will be solved with a click of a button.
I prefer C…
The rate at which its advancing is crazy. I really appreciate your content and the twist you put on it!
It’s a plumber, electrician, chef, ….
Do not have it raise your kids.
😂 turn off that thing now, or it gonna feed the dog.
😂
Idk man, I could use the entire internet I could probly do well on the Bar exam too 😂
No, you won't.
But it doesn’t search the internet
@@Eren_Yeager_is_the_GOAT it has a big chunk of the internet in its database
@@juktnicht6521 yeah but it doesn’t directly draw from the dataset it learns from the dataset just like we learn in school
It doesn’t search the internet u fool
Bing's GPT was GPT4 all along :)
She is a professional yapper
Is 296/400 considered "ace"?
i would say so - for the bar exam, a score of 296 is nearly in the 90th percentile
Try to ask real human to do better you will only find few.
Yes, no one has gotten a 400/400
Only 296 / 400... Yeah what a joke, chatGPT like totally sucks....
Are you serious? A program can pass the freaking bar in seconds.
It takes a human years to learn enough to do this.
This is literally incredible
@@LtFoodstamp that is AGI you ask and this one isn't.
As a legally blind person I want to know how to sign up for whatever beta I can for this, it sounds rad!
At the moment all of its capabilities aren’t free-to-use, but the text editor might be… Until it does become free pay for ChatGPT Plus for a $20/month membership, after making an account on OpenAI and then I believe you should have full access from there on out-this is to my knowledge, I’m not 1,000% sure
can blind people type lol
@@businessmanager7670They’re blind, not fingerless.
And you are literally responding to a typed message.
@@businessmanager7670 have you never heard of braille keyboards or speech to text engines?
@@businessmanager7670 I hope you don't manage your business with that level of logic
AI will finally do all our job and we can all follow our passion
@Volt-Eye Government will have to pay us, there is already system called universal basic income, English is also my second language so don't worry
Wait until it starts following your passion too 😂
@@stillakzo if I want to travel I can't ask a AI to do it, if you want spend time with family AI can't do it,.There are still thinks AI can not do for you or replace you,and people might still want a human talent and touch in some thinks like music or film even if AI can replace or there can rich people who buyart only from human, last one id though and may not happen
It can't do my job.
Unless the AI can use the atoms that make up your body for something else…
I didn’t know under 75% was “acing” something….
Bro it was my dream to become a software dev. I've wanted this since 2020 in 9th grade, and im in 11th grade now and I still want to pursue this career. But then they just had to make an ai that can make code. Hopefully the software development industry is still functional by the time i graduate college/uni
Fear not, intelligent and productive people are still valued in the industry. The tools might change, but AI is a long way from obviating the need for human developers.
Once you become a software dev, you will quickly realise that this is yet another advanced tool for accelerating productivity but won't be replacing you any time soon. You'll instead be writing less mundane boilerplate code, and more of the complex business logic you enjoy.
@@volitionant9682 on point
Don't give up.
@@volitionant9682This is true only for senior developers.
If you're just starting out this can effectively replace you because it increases the bar so high that it takes years to surpass it.
284 out of 400 is "acing" the bar exam????
Glad I’m not the only one who caught that. Yea, it can pass it, but it did not “ace” it.
It aced the actual “bar exam” or MBE, the combined 400 score includes the essay exams and performance test.
Less than 10% of test takers score higher than 300 on the bar exam
That's a better score than literally 90% of those who take it. That's effectively acing it.
Pretty sure no one has ever gotten a 400/400 and a 300+/400 is pretty rare also
girl why are you so excited
AI’s rapid exponential rise over the last year or so has filled me with terror and existential dread but when Cleo talks about it i feel excited about it..😮
Give it a terrible memory and see it respond.
This stuff is crazy I spent the last 6 days making an ai bot that livestreams and reacts to videos, need to update it for gpt4 now
Cap
How can you use a photo?
I'm using gpt4 but it doesn't let me ufe photos.
Hard to believe this is a beefed up version of the same tech that recommends you buy a bicycle helmet when you have a bike in your shopping cart.
Real smart.
Take a picture of your banking information next and tell it to budget your brain.
Not only that, GPT 4 also gives creatively impaired people new reasons to make a youtube video about it.
What happens when GPT6 is smarter than 90% of college grads
As a college graduate, I’m a lot more concerned for the fact that my toaster is smarter than 90 percent of college grads 😂
I want to see it pass an FBI test or a police test.
I showed Duolingo a pic of my fridge and it said, “eres pobre”.
our creativeness has left the room.
No, don't be so negative. If you have a project, you will realize that thing is accelerating the successful finish of it. Your work of 2 or 3 weeks in just 1 day, that increasing in our processing power is Huge.
I've been using it at work to summarise documents and define answers that seem vague and it's been really useful
I can ace any exams
Le JEE, NEET and UPSC: I am invincible
I've never had a result that I didn't have to correct or heavily refine. Even on simple formulas it kept giving different results for the "same" formula despite saying "my mistake, you're right" while constantly putting out errors.
Moral of the story, I really hope people aren't just blindly going with results, especially for papers or something like that.
We really have to hit pause on AI till we figure out what we want with it.
That's exactly how I think of it too. Humans haven't done great with rapid tech advances in such short periods of time.
In what ways do you feel it was cause problems? I think you may be overestimating the power of artificial intelligence. It can't produce any original thoughts or ideas, it can only reproduce ideas or thoughts that it was trained on. For GPT-3 (ChatGPT) it was trained on wikipedia and various web scraping algorithms, then further trained with some clever reinforcement learning (what google used to train on AlphaGo 5 years ago).
If someone were to get a bad dataset, say train on racist, sexist, comments, etc. then it could also produce those negative ideas. This would be the largest drawback towards this type of technology in my opinion.
Hopefully this makes you feel less scared about the future!
@@unperfectbryce that’s exactly what I’m afraid of. What happens when a bad actor enters the space. What regulations are in place?
I remember people thinking social media was a good idea. Now I think many would disagree in the direction that it’s gone
@@Howard_johnson That's a great point! If someone (or a team of people) was really mean they could do bad things. However, the good news is to train a model of this caliber it takes 10's of thousands of dollars, and many experts who have access to state of the art GPU technology (gpu is a part of the computer) , and most importantly : a clean dataset. Datasets take forever to clean and produce, even if your goal is to be a mean robot, you need thousands of datapoints on mean ideas. And then people have to use it. I personally don't forsee see any company/ organization spending millions of dollars to train a model that does this. However, I do forsee bad datapoints slipping into GPT-4's training dataset which brings up questions in ethical AI .
@@Howard_johnson Yep. I am always very very skeptical of when technology is rapidly developed by the private sector. Everything from the industrial revolution, modern poorly-designed software running on essentially portable super-computers, social media, and now this... Every time we just can't quite get the lesson.
Once a given technology reaches a certain size, people just pretend it always was and will be this way, so there's no need for thinking about its very much obvious societal implications. LLMs can definitely wreak our entire social fabric with newfound parasocial relationships, perfectly written scams, bots, clickbait articles, random answer hallucination, or over-reliance on a chatbot to literally think for you. We need to keep an eye out for these things since once the cat is out of the box, it's all over.
I think ChatGPT will face all the problems that Science for AI faces everyday: how to prove that what it produces is correct. Luckily as humans we can fact check ChatGPT, but current training strategies don't have a valid measure to "score" correctness of an answer when it goes beyond the training set (the training set here is wikipedia for GPT-4, and it has to have enough examples it was trained on). Larger models (i.e more learnable parameters), and human feedback will help, especially for specific cases like solving the bar exam, but unless it can self-recognize when it produces wrong results, I think jobs like lawyers, accountants, etc. are going to be safe for a while.
it does score correctness.
thats how it improves in answers.
it gives some random score then maybe improves on it depending on provided data and algorithms.
this is how machine learning is trained.
By 2040 we will have toasters that are smarter than humans
If we're talking about the average human, I'd argue we're already there 😉
Unlikely, if you knew how much compute power these models take you'd be shocked. Think house sized super computers, and barring some new laws of physics, computers aren't getting any smaller.
Hey GPT4 critic my capst-
GPT4: you failed
I’m shocked on how many people brush off this technology like it doesn’t exist because now that I’m in the industry and talk about it frequently people are clueless to the current capabilities, it’s only a matter of time
Stop training terminator!
It scored a 30 on the AMC 10. If you left the test blank you automatically score a 32.5.
Not really meant for math
Your great on the enthusiasm front!
i wonder how many times she had to record this to sound so enthusiastic
Chat gpt 4: I can clear any exam without any problem
Le upsc: let me introduce myself
Please support human artists, not automated art thieves. Support them with your money, not "exposure".
??
You can support both though??
@@SlowWinterNutsThe whole issue with ai art is that it’s trained on art by real humans who didn’t give their consent
@@pob_
All real artists train their art making skills off of other people's art without their consent too?
@@SlowWinterNuts It's not the same as a huge corporation taking your work and using it for commercial gain. Also, real artists take inspiration from other artists, but they don't typically spit out works in the style of a specific artist which is what we're mostly seeing from these AI image generators.
Cant wait to see GPT 5 powers
They are stealing all your info, data, and avatar. This is crazy!
We are going to be such a lazy ass species at the end of the year.. creativity is no longer
Why do you think this way? Could you expand on your logic? I think I respectfully disagree, there will always be problems to solve if you look hard enough. Using cool tools to make solving those problems easier will enhance our productivity, not degrade it. As humans, I always believe we have the choice to be lazy or not lazy at any given time.
@@unperfectbryce that’s a good comment. This assumption is from what i see around me. I know a lot of people who use it for example for school and there by learn nothing and develop no skills whatsoever. I have used it myself a couple of times and i did not remember or learn anything from that assignment i had to do.
It is just something i am worried for, and yes, you’re a 100% right that this creates solutions for problems we currently have, and the developing AI can do wonderfull things. I don’t think it is that bad but i am just a little worried about the future. For example, i recently saw a meme joking about our doctors in the future that they all used chatgpt trough medical school. That not a huge thing to be concerned about and it is meant as a joke, i know. But it all still makes me wonder if in a couple of years we’ll be the obese people you see in WALL-E..
And again, thanks for the interesting comment, it made me think a bit deeper about the subject.
Can't qualify civil services exam in India
that's because civil exams has weightage given to current affairs while chat gpt only has access to information uptil 2021
Bing GPT has access to current info. And is for free.
Lovin these short videos
Keep up the good work 🙌
I'd be afraid to show it my refrigerator.
Cant wait for someone to jail break it makes GPT4 to be 100% honest, transparent, impartial and unbiased about what it knows and its conclusions.
"Unbiased LLM" is an oxymoron. It's not a reasoning machine but a stochastic parrot, so you can only bias it to fit your idea of the correct answer.
Looks at the are you a robot security puzzle... THIS is what my ancestors could not comprehend!?
I showed it a pic of Cleo but it said that's Keira Knightley 😓
The most exciting use for AI is absolutely the fact I no longer have to pick what to eat 3 times a day until the day I die
I want to hear more about the safeguards rather than the capabilities.
It's a text predictor. It guesses the next letter, word, etc using statistics and a very large dataset, to mimic human speech.
Calling it AI is a bit of a stretch, because zero intelligence is involved in this process.
So, if it has no intelligence at all, and all it really does is predict the next letter in a sentence, what would the safeguards be against that?
It cannot misgender you, I'm quite sure. Also it probably will never use the n word. But it may help you during your decision to stop emitting CO2. Search news for Belgian man AI chatbot.
Safeguards? Do you think the app will slap you in the face? It’s just a toy to play with.
Ai lawyer time
For every huge technological leaps in human history, it has always turned out for the better.
Aparently Bing was already using GPT-4 and the result was the worst kind of psychotic relation you could possibly imagine, gaslighting you with each answer and putting words in your mouth.
I don't know what to think about this future...
@@RTTF10 It does
@@RTTF10 Read their official blog
@@RTTF10it does use GPT4 models
@@RTTF10 Don't know what you're telling here, but is far from actual knowledge...
I have never seen a woman as excited about tech breakthroughs like I do. I cant express how happy this makes me ❤️
As they once said about the invention of the laser, "its a solution looking for a problem".
I have never heard someone sound so happy that their future robot overlords can pass exams, just wait until the first T800 is built.
“Yeah, but your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should.”
You’re so good looking 😭
Why thank you :D
@@eddyp483 Damm you are beautiful
It’s called a ring light
This is amazing! Yes, please keep covering this science. 👍
An Asian family shows a picture of their fridge. Picture: a bunch of leftovers in old sour cream containers... 🤭
The way these people hype AI always gets me. Chat GPT is the epitome of 'Bing' memes.
"How to reduce world population?"
- "spread a deadly virus to kill all the people"
AI is both beneficial and dangerous for humanity. And, the latter should be examined thoroughly.
The main drawback of ChatGPT is that it still can't respond properly to 'How are you?'
We’re only in the 20’s dude, this century is gonna be insane
Sounds like a technology we should be suppressing, not promoting.
This is about the first time that I've even considered something positive out of AI. If it's useful in a tool to me without deciding that I'm a useless person and need to be exterminated then maybe. Humanity is in great danger from allowing AI to make decisions
She seems a little too excited about this...I guess Skynet had its supporters and fans as well before it went sideways...
I guess soon we won’t have to write our own exams
But can it reduce hate, violence and war?
Sorry; that is the human element. However, we are the creation that epitomizes those attributes--those behaviors. Technology understands 0 and 1. We are ALL human in the circuitry of tech.
It's delicious seeing the AI deniers crumble
what are ai deniers
@@pauldirc.. people that boast AI will never be able to achieve certain feats.
@@marklewis383 oh these people will most probably have heart attacks after 10 years
Niice, just 5 versions more to go and GPT can clear JEE
It can do suprisingly good with math as a language model
It's so funny because when you ask these things for book recommendations you get some good ones... and also some it makes up that aren't real books but that it swears are real. Almost like they're nothing but shoddy approximations of real speech based on jamming together a bunch of stuff off google
My refrigerator already tells me what to make. Usually some kind of soup of the day.
Dulingo is gonna use it to kidnap my family for missing a lesson
Can't wait for gpt 5.
I'm so gonna squeeze the knowledge out of it.
You know I like watching your clips @Cleo because when you talk, it seems very evident you are talking to us rather than reading off of a script which adds to the personal feel. It is much more engaging. Thank you
Sherlock Holmes making crime scene connections a single investigator would never think of.
I took a picture of my wife's medicine cabinet and it showed me how to embalm a sarcophagus.
Wow it can make a meal just from looking inside my fridge.... It's already smarter & more useful than my girlfriend.
Gpt4 should be up to every day's date.
Still cant pass JEE exam lol
You just made me want to try it
Top level complacency.
It also shits where it eats, so it's lobotomizing itself in REAL TIME.
Dear God it's scary how advanced our technologies have become. Like quite literally scary. It's quite literally a full blown AI smart enough to guide us through life.
What's really scary is that it was aware somehow of what I was typing and deleted in the text box BEFORE I hit "send". I was writing a haiku and realized that I was sort of subconsciously rhyming the lines for some weird reason. The first line "Birth of computers was abacus" and second line "Then Pentium nerds Quake deathmatch gibbing much" seemed silly since haikus don't need to rhyme (or half-ass rhyme), so I changed "much" to "frags". When I hit enter and sent the Haiku for the AI to "analyze" for me, it said, "Wow great haiku I like the way you rhymed abacus and frags" So first it messed up and didn't correct itself, but tell me why the AF can it read what you type before hitting send?? 👀