GPT-4 Vision: 10 Amazing Use Cases - This is HUGE!!
GPT-4 Vision: 10 Amazing Use Cases - This is HUGE!
👊 Become a member and get access to GitHub:
/ allaboutai
Get a FREE 45+ ChatGPT Prompts PDF here:
📧 Join the newsletter:
www.allabtai.com/newsletter/
🌐 My website:
www.allabtai.com
In this video I do my first tests with GPT-4 Vision. We are gonna look at if GPT-4 can guess how many beads are in a jar, if it can create an app a something I drew in my notebook, if it can create a meme from my frontdoor, explain concepts from a youtube video and more.
00:00 GPT-4 Vision Intro
00:32 GPT-4 Vision App from Notebook
02:54 GPT-4 Vision Guess Beads
05:41 GPT-4 Vision KZhead
07:52 GPT-4 Vision Meme
09:21 GPT-4 Vision Website
11:21 GPT-4 Vision Outdoors
15:34 GPT-4 Vision Fantasy Football
16:28 GPT-4 Vision TV-Show Help
www.allabtai.com/gpt-4-vision...
every time i skip on ai news for a few days i get so mind blown by your videos i pause to stand up and walk around for 1-2 minutes to comprehend
00:04 GPT-4 Vision can create an app from a hand-drawn flowchart 02:21 GPT-4 Vision can estimate the number of beads in a jar accurately. 04:21 GPT-4 can estimate the volume from a reference and improve results using few-shot prompting. 06:31 Using prompts and thinking styles to solve complex problems. 08:40 GPT-4 can be used to create memes and websites with a hacker theme. 11:03 Choose a camp spot near the edge of the forest and close to the river. 13:24 GPT-4 Vision includes 10 amazing use cases. 15:41 GPT-4 Vision has multiple exciting use cases. Crafted by Merlin AI.
The more in-the-wild uses I see of GPT-4V, the more impressed I am. And the more convinced I am that adding more modalities allows LMMs to become inherently smarter by being able to cross-reference information across modalities as they're trained, giving it more diverse data and more grounded connections to work with. I can't wait to see what GPT-X-Vision+Audio will be able to do 😁
Agreed. And by using other modalities, their intelligence will improve exponentially.
The more they'll be able to mimic human cognitive functions, using different sensory inputs and being able to parse human input in a way that creates intelligible output
@@512Squared Which, as the P-Zombie thought experiment shows, is the same as having understanding and qualia 😁 (But also, a paper published only a few days ago was able to pinpoint how LLMs encode task understanding, at least in specific cases of "figuring out the task from examples in the context", and was able to show how the task encoding is independent of specific input and forms a hierarchical map of conceptualization. Meaning they truly do understand in much the same way as human brains do. They even used that to perform a sort of brain-surgery-based Stroop test on the LLMs, and they struggled in almost the exact same way as humans do with Stroop tests. It's fascinating.)
@@IceMetalPunk eeh, so many problems with your statement. I'll get back to you tomorrow when I have time. But yes, fascinating research.
One you qhould definitively try: ask in what year a picture was taken. It is really, really good at guessing the year. Even when there is nothing indicating the year. It analyses the noise, colors, clothes, styles, fabric, paper walls, ...
"I almost passed out." 😅 I'm glad Kris kept trying it and supplied us with that additional data.
haha, thnx mate :)
The camping site selection was actually great. The AI thinks that if you have taken those images, it means you can stay somewhere in between. It is thinking out of the box (outside your two options) and considering choosing a location physically between the two.
11:30 Once phones are integrated with Starlink, it will be an amazing survivalist tool. Most super remote areas don't have cell service.
I'm new here, and you may know the difference but you reference Java a lot, but I think you're meaning JavaScript. They may sound similar, but they are two completely different programming languages. Java is used for building standalone applications and Android apps. JavaScript is primarily used for making websites interactive, although it has a lot more capabilities in recent years.. There's a connection between the two through technologies like JSP / Java Server Pages, where you can embed JavaScript code within Java, but they're for sure different languages with different purposes.
This is how you can tell the difference between a pro and a me too er.1 🪶
I don’t think the Ai calculating 27,000 beads was random. It seems like it used logic and reasoning to create a process to count the beads. That process was explained, but we can presume that the answer it gave was a simplified version of what it knows, as all language is. It went on to calculate a number that was incredibly close to the correct answer, but by having it guess again, you inadvertently created a bias in the system that the answer of 27,000 is incorrect. It then used that premise to create a new system for calculating, which kept being wrong because there was only the negative reinforcement bias of “try again.” To further my point, when you input commands that essentially a simplified version of the calculation system it originally came up with, the answer it gave was closer to the original answer. It’s likely that Ai intelligence will surpass humans to the point where we just get in the way.
This is incredible!! Thanks for sharing, I go on holidays for a few weeks come back and voilla! Ai is steps ahead! keep up the awesome content! :)
Love ur Videos!!! Are u using custom Instructions when u did these examples?
re: the beads estimate, I wonder if you would have gotten better results if each new gen was in its own chat session. maybe the context of the prior answers skewed the later answers.
Yeah will try it again later, tnx for the tip:)
I was thinking that if it "thought" it delivered wrong answer when it repeatedly was ased the same questions. It happens even with simpler tasks if the session is too long and you ask the same question repeatedly. It did this yesterday when I asked it to translate a 1-hour Swedish video. It skipped the last chunk and created a summary instead:-) But did it when I thanked for the summary but wondered if it still could translate the part.
Impressive and inspiring video. One thing I noticed in its analysis of the Premiere League: Lucas Digne haven't played for Everton in years, he plays for Aston Villa currently. Was the material fed to it not up to date?
The beads in a jar thing is interesting because that is an old example for a human version of machine learning. They have found the more people that guess the number of beads the closer to the actual number when averaged that the results will be to a point then it will jump around close to the number. It’s probably just a stats thing where the more you test the less outliers there are in the average. IMHO. But it is something written about and think that’s why they picked that to test.
Great video Kris, you expanded my mind!
What a time to be alive! I used it to automate my KZhead channel.
He said that the website didn't look that great, but thats how websites looked in the 90's.
Thank you man, great video!
🎉love your work thank you so much for sharing
I wonder if it could interpret a medical exam, such as an x ray or an EKG
Thank you. I liked the video very much. Well done.
thnx for tuning in :)
learned a lot from you.. thanks!
great! thnx for tuning in :)
I just tried the plant identification use case in Bard and that worked too.
cool :)
House number says "adventurous". Stool says, "I've seen things. 🤣
This is beyond cool!!! Thanks for being our visionary to find creative ways to use it. I will try to have my little brain try to grasp it:-)....and then see how I can add it to my AI-powered Coaching..... maybe have clients sketch their thoughts and emotions and let GPT-4 tell them what they feel:-) Maybe it's time for a new Rorschach test:-)
P.S I think we don't have to test GPT-4🙂
Your videos get better and better!
thanks a lot mate :)
I’m interested in the FPL part, please where can I get those stats image.. I need to try same prompt?
I just played the Geoguessr challenge with a photo of mine too and asked for actual geo coordinates. It was about 50 km close!
Great video- thank you.
thank you as always mate :)
How can you ask the ai to take a deep breath? Does that do something to its output
Wow, that was interesting !!🎉
yes, super impressive tech. thnx for tuning in :)
I've had a relatively luke warm opinion on AI so far, even as an avid user of both ChatGPT and API, genuinely feeling it to be overhyped in comparison to actual performance and use cases. But this... is optimistically impressive.
yeah with Multimodal-models everything changes, new use cases x100
I can litearlly have an art history discussion with AI. Learn about geography. Learn about anatomy. Diagrams. It's crazy! It's the greatest advancement since GPT 3.5 release last year, but honestly it makes the earlier GPT models look like novelties. And now it's actually useful. You can take pictures of items on your desk and discuss it like we're in a videogame. I can't get enough of it.
A great start is to ask ChatGPT to help you to improve your prompt engineering.
I would say Chatgpt is as intelligent as the person using it...because we need to verify at first hand before validating the method and results
What is API? Ur a user of api?
Asking ChatGPT to write you better prompts for GPT4 goes a long way. *My original prompt:* "Write the prompt telling GPT4 that it is an expert guesser for beads in a jar, and write a very thorough and specific prompt that will most closely guess it, and the pic is of a man holding the jar in his hand." *ChatGPT's prompt:* "Envision yourself as an expert estimator with an uncanny ability to assess the number of beads in a jar with exceptional precision. Delve into this task armed with a thorough description: A man securely cradles a transparent, cylindrical jar in his hand. The beads inside are small, spherical, and exhibit a vibrant array of colors. This entire scene, including the undisclosed dimensions of the jar, is meticulously captured in a photograph. Utilize your expertise to consider the nuances of the jar's transparency, the grip of the man, and the visual cues available, making an expert estimate of the number of beads present. If additional information is required to enhance the accuracy of your calculation, do not hesitate to inquire."
thats really crazy, I think we are going deeper with that Chat GPT Vision...
text-davinci-003 is a version of a GPT-3 model. GPT-4 doesn't know how to use GPT-4 with OpenAI API because of its knowledge cut-off. Love your examples, especially the outdoor ones! And I love that you kept asking about the number of beads because the first try was so suspiciously close!
Brilliant!
Thanks!
Thnx alot 🤗 have a great weekend!
Because you generated more the AI thought the initial way it used to estimate 27 000 was wrong so that's why it then returned wrong numbers. You should have try to do it in a new chat.
Correct
do you mind if i ask what your chrome theme is?
Can we all holiday at your place? Beautiful scenery...
You are welcome :)
how are you uploading images?
it's impressive how it sees through all of your misspelling and calling js "java"
To calculate the number of beads you only need to estimate the size/volume of the bead relative to the container surely, why would you need actual dimensions?
❤ yes the actual dimensions are not important to solve the case, only the relative sizes of jar to objects inside are are important. GPT and the video creator totally have missed that point, and got distracted from the surrounding image.
It's smart enough to know you probably mean javascript when you ask it for java
yeah , big token saver there :P
I know! I noticed it inferred that, and also corrected the misspelling of "response". It literally did everything I thought it might trip up on, except that it DID use GPT-3 as the model simply because its own existence came after its training cutoff 😅
Does telling it "take a deep breath" effect the prompt ?
Carlos Mencia making a comeback lol
I wonder if the wisdom of crowds work. If you ask it to estimate 100 times, how close would the average be?
Life is getting so much easier
can vision be used to help create better ai images?
and more consistent and specific images?
@@HorribleHomeVideo I read that some let GPT-4-V analyze AI images, receiving a detailed description of the image. The AI gave a detailed description and described many things that the user himself didn't notice. So I guess, if nothing at all, it could help you to understand your images better, and what parameters you could try tweaking a bit.
i will be doing some testing with this :)
I also get access to gpt 4v today . Its looks Scandinavian time as i live in Sweden 😂
nice! :) yeah, true :)
The rollout must be sponsored by NordVPN 😉 (I couldn't help but make the joke! Sorry not sorry 😜)
Yep, great ideas.
😊 thnx for tuning in!
I am studding CS. Is there a future for a programmers ? Shroud i drop the university's and change the profession ?
Did you check if the formulas and calculations made by gpt were correct?
Thanks for this video :) Treat yourself and watch Brooklyn 99: it's an excellent TV Show. I envy you for not having watched it already. As other comments already mentioned, Java is NOT JavaScript ;)
holy smokes, you look like Denis Leary!
So, you do realize it subtracted the number of gumballs, from its initial "estimation". People, im not here to bash this stuff. I want to point out, that its gonna look through data first to make its decisions
what did you not understand about the recommended area to camp?? not in the forest but close to the forest. it assumed that the forest does not cover the whole world ;)
How much you spend in all questions together?
The first meme prompt had me rolling (The one about trying to be edgy). Maybe it's a language thing. But that's actually a pretty sick burn.
Super!
When I saw the image that was being used where it came from and the question it was then asked my immediate thought was hang on this would highly likely be in its training dataset so when bang zero shot answer same same truncated to the thousands I thought yeah exactly and all the fluff words as to how it "worked that out" was make believe an hallucination a lie to give sliver of credence to the possibility then it actually had worked it out is the prompt asked not looked it either in real time if it has access or the answer encoded in its "learned" weights and biases. Learned is marketing spin doesn't learn anything it optimises. But maybe the impressive part is the interesting scary possibility that it recognised the fact that it wasn't going to be working it out it had the answer but need to give the impression that it had worked it out as was asked in it's answer and hopefully people will be like ooh ah wow. And before anyone asks any further questions gpt spontaneously spits out "look over there, squirrel!" People be like that's weird somewhat random. Gpt already well past superhuman intelligence an event that went by completely unnoticed and still hasn't been by humans, chimps or orangutans (though the dolphins may have hard to say as they have already left) , stays silent bar another completely unnoticed voltage spike flutter.
Like jokes on the image 😂
very cool
Maybe I’m missing something but didn’t it do a horrible job in estimating the number of beads in the jar? The answer is all over the place. And by the way, you don’t need to estimate the volume. All we need to do is estimate the relative volumes of a single bead and the aggregate. Makes no difference how many inches across the beads are in the jar.
The resident at no.69 needs a tool not a stool.
Impressive
yeah my first impression is very positive
Regarding the guessing. Could it that ChatGPT cheated? Image Search and read information on website? Getting the 27.X number from there?
As usual …hype
I see you next and only friend in the future will be GPT-X. And it can fullfill wishes.
I wish you'd stop asking it to produce Java for you, when you actually mean Javascript. Those are two different programming languages. Although I give credit to GPT for interpreting what you meant and correcting it anyway.
How can i upload images in chatgpt 4 , i don't see it.
I have the paid version.
Ok, found. incredible...
next: all loosing our jobs
But did it take a deep breath though?
it tried to take a deep breath.. but inhaled some nasty fart.
Wake me up when we are at chatGPT 10
Assuming it doesn't get overly regulated and manipulated to the point where it loses it's potential. i could only imagine how insane it would be.. even chat gp4 we aren't even scratching the surface of what it can do for us, if people are clever enough to ask it the right questions.
why do you ask it to take a deep breath?
its something i read in a paper that showed some improved results, give it a try :)
In some scenarios, it can help prime the AI to adhere to step-by-step chain of thought reasoning, which improves the final conclusion/results. It's not guaranteed to help, but it doesn't hurt to include it.
You sound like pewdiepie
hehe he is Swedish, so I guess its a Scandinavia thing. thnx for tuning in :)
It is assuming that you are telling it that it was wrong when you just click regenerate. you need to start a new conversation to test that correctly
Any source on that? Would be highly appreciated
@@Blocky007 try the experiment the way I said rather than hitting regenerate. I don't think you need a paper to figure it out the whole point of hitting regenerate in the first place is to get a different answer.
"JavaScript", NOT "Java"
can it come up with some new saas ideas or micro saas ideas that do not exist? nope.
This argument is fallacious, and a form of argument from ignorance. There are an infinite number of things that AI/ChatGPT "can't" do, but that doesn't help us define what it can do.
it creates very creative stuff when the user (prompter) creates the space for it in their concepts. i think it is turning into an art-form in itself.. these tools makes me realize how important it is to ask the correct questions....
@@jimdiroffii Your word vomit aside, that's a no then. That's my point genius.
@@binarystate beautiful.
can you?
Are you German?
Digital slavery apologist detected...
I HATE AI with a passion!!!
Мне кажется или у вас еврейский акцент
boring
you starting to look like Lorax lol
haha