Thanks to our sponsor Abacus AI. Try their new ChatLLM platform here: chatllm.abacus.ai/?token=aisearch Prompts used in the video: create a snake game using python Create a single HTML page that let's me upload an audio file. Sync that audio with a visualizer like the attached image. Don't use unsupported libraries. convert this UI design into frontend code create tetris game using python create a JS presentation on the health implications of coffee Create an interactive 2-page infographic on the attached document Using the neural network diagram I've shared as a reference, please create an interactive HTML/JavaScript animation that demonstrates the flow of data through this network. The animation should include: 1. A visual representation of the network structure, matching the layout in the image. 2. Animated paths showing data flowing from the input layer through the hidden layers to the output layer. 3. The ability to input sample data into the 5 input nodes. 4. Visual feedback showing how the activation of nodes changes based on the input. 5. A simple UI to control the animation speed and reset the simulation. The animation should be educational, helping users understand how data propagates through a neural network. Please provide the full HTML and JavaScript code needed to run this animation in a web browser. create an app in one html page that can be used in Artifacts. make an interactive 3D particle cloud, with a max of 100 particles. use Three.js for the simulation. do not use unsupported or 3rd party libraries or functions. create your own functions.
I've been a ChatGPT boy, but your video made me try to experience Anthropic but the UI that I have is quite different to the one that presented in the video, specially it does not have the cool preview option for JS or HTML, is there anything that I am missing ?
Don't skip the first 10 seconds. I spent hours trying to get to that stupid slider switch, which, once you've started a conversation, you cannot enable. )(($*$*)_#)_@)_!
Well, the funny thing about the Snake game is that it created the exact same game for me as well, with the same message and everything. It seems to have been trained specifically on that, or for all we know, it could be identical code pulled from a database.
@@theAIsearch I've been wondering whether it's pulling from some database or if it's just incredibly consistent. Remember that other AI guy, Matthew? (I can't recall his name exactly, but it starts with an M.) He does a lot of testing with snake games too. I thought it was an interesting observation. It seems like the AI might be highly trained on the snake game by now, or it's drawing from a database. It would be interesting to run many different trials and see the exact code that's varying and changing. This could provide insight into what's actually working under the hood. For all we know, it could be utilizing a vast database of open-source code, grabbing that code, and using an AI in the background to modify it. If they aren't doing that, I'd be disappointed because it would be a great idea.
I noticed in the coffee presentation nothing was said about polyphenols or all-cause-mortality,... Maybe if Claude 3.5 was told to create a graduate level presentation or a presentation to researchers? So far I've been extremely happy with Claude Sonnet.
Absolutely love that it nailed the entire snake game but could not horizontally and vertically center the message when you run into a wall or yourself 😅
CSS or fronted is hard as f* There are too many options and can conflict with each other. And AI can't try it out. With ChatGPT this is the thing that I override. It works best when you use frameworks.
@@hornerfarah2282 By "hurt" you mean indie devs and regular people could create well made games in less time with less experience and it would overall be better for everyone besides AAA studios that milk terrible products every year for money how horrible
@@ProfShibe To be fair, after all of these AI's getting better and better, I feel that studying programming languages is going to become worthless, it's discouraging me a lot these days
@@darkkittenyt6001 As long as clients are as stupid as they are today, programmers will still be needed. Half of the time they can't explain what they actually want, so they won't get the wanted results from an LLM either. Even if they get a working application out of it, they can't debug issues, scale the codebase or keep track of security flaws. Let them try, they'll come crawling back soon enough.
@@theAIsearch Well even the paid version is quite limited if you use it a lot. I mean I'm trying to do a kind of an advanced .php site with it, so I've hit the limit like 3 times today already. If not those limits, then the sky would be the limit lol
@@theAIsearch I get around 4 messages every 3 hours. According to Anthropic if I upgrade, I will get 5x tokens. So, are they saying I'll be stuck at 20 messages per 4 hours or 25 messages every 5 hours in your case? Seems very restrictive.
@@orthodox_gentleman Good to know. I suspected that may be the case. As someone who literally engages in occasional sessions that are hours long, this does not sound optimal for me.
I just signed up based on this recommendation but my experience was different. Won't generate pictures, work with spreadsheets, etc. Maybe I missed something? " I'm an AI language model designed to analyze and describe images, but I don't have the capability to produce or modify visual content." "I apologize for the confusion, but I cannot directly create, generate, or manipulate any kind of images, including spreadsheet graphs. I'm an AI language model designed to process and generate text-based information."
Well it can't really create visual, as far as I understand it. Just code. For the many examples he copied the code to another program to run it, and for the others he had it use a web Server.
I made a 2d fishing game with it. So you click a button called fish and it takes a second and you either collect a clam or fish or crab and you gain money also if you buy anything you get coolness as I added a coolness meter as a joke. Also if you have over 50 dollars you turn yellow and over 100 you turn blue and over 500 turn white and over 1000 neon red. Also a little meter shows when you and what your next rank is. The shop holds 3 upgrades the first is called extra hook the 2nd is extra cash and the 3rd is faster fishing.
Just got it to do a five page interactive document on the timeline of FNAF and it actually helped clear up the story a little more, however it didn't go into any detail regaurding the books as i only specified the video games.
I used to think gpt4 was impressive, but then i discovered Claude 2.0. I haven't looked back since - Claude flat out destroys gpt4 in all aspects i use it for. When 3.5 released i was mind blown by how much it improved further. Can't wait to see the next iteration - if GPT5 isn't going to be x2 better than gpt4, they just won't be able to catch up no matter what they do.
My experience differed wildly, it was very crude. Failed to make, draw, or reproduce a simple visual representation of an electronic schematic when I provided an image.
I have used both and to tell you the truth, i really cant pick just one. GPT is great when it comes to figuring out what to do when problems occur with complicated business logic that involves code from various places, I couldn't get Claude to figure out my problem that easily, its solutions did not work for me. On the other hand, I was facing a very specific problem with local notifications and GPT really could not help at all, like it immediately started repeating the code after some attempts. Claude managed to help me step by step very methodically. I really don't know, maybe it is on me but I really can't choose. I need them both
The music is a great example of ais limitations. The Leigh man might think that song was amazing but music to those who create is a much more expressive experience. Pop music has become so formulaic that our most simple iterations of ai can recreate top chart songs in an instant or new ones because music has regressed so much the past 20 years. Even if you had bots inside of the most accurate human robot it would never be able to recreate the live version of machine gun by Jimi Hendrix with the same level of emotional projection. This ai is super cool tho, on the working side of things this ai is ridiculous.
Having Patrick star's head explode is a pretty good example of what's going on in my head right now? Did it really make a video game? 🤯 Where did this come all the sudden?
I always knew that the ais where purposely lacking due to coporate being corporate because when you really think about it this is just slightly above what ai could actually do at the time
All the LLMs still fail to process image processing. I just made a simple request of adding a hand to an image like holding the given image, and all 3 big LLMs failed (Gemini, Chatgpt-40, and Calude 3.5)
Switched my subscription over from GPT to Claude. GPT is pretty good, but I hate how inconsistent its quality can be. One time it's amazing, the other time I use it it's dumb as bricks. Having tried Claude for a bit, it's definitely a step up over GPT in overall quality, at least for writing code for my game.
I used it to integrate a service API it did it in less than 3 minutes, and the service works exactly as intended, The same prompt was used with ChatGPT gave me exceptions and errors
I do notice it has allot of limitations too. Try creating an 'liquid with gravity' using particles inside half a square. that the user is able to rotate and then the particles fall out. its too hard. I get alot of bugs and glitches.
8:50 I compared the actual spectrogram of the song, and whilst it already has a terrible mixdown and quality, the visualizer just portrays the spectrum entirely wrong. On the far left side, we'd expect the lowest frequency range starting from ~20Hz. The song has not a single signal in that range. The mids and highs are _much_ louder than what this visualizer shows. Whilst I am impressed of this AI producing valid code, I am not impressed by the result.
Destroys this destroys that… and next week there is another update for ChatGPT and all functions are included there too. Really not every little thing should be blown up to „this ends… this destroys… forever….INSANE!!!1!!1“ stop making such titles, because at some point people won’t fall for these superlatives in every title anymore, at least I won’t.
this is actually scary how it is able to do this ai has already taken so many graphic design jobs next its going to take game development and programming?
I think this is very cool but we can’t let kids get a hold of this sort of technology at a young age. They are going to lose the ability to think on their own and to critically think
Intellectual property laws will joyfully be under pressure. Power will shift and there will be commercial chaos! The outcome could be ugly or there could be real equalization. The only constant will be change. Say you use an AI to create a new game based on an idea you prompt for. Then you publish it for sale and it becomes a hit. How do you think the intellectual property ownership should and would be determined? Who owns the judgment? My view is that IP should have had scale limits and forces may finally see that limits need to be set in IP law. Creativity is a shared existence.
The only problem I have with it is that as long as it doesn’t have some kind of voice mode, I just wont use it as much. It’s still exciting to see the development and progress though.
I primarily use Chat GPT for image creation. How does Claude compare? I can get fairly consistent and accurate to prompt new characters from Dalle. Most the Art generators don't hone in on a unique art style with a simple prompt the way Dalle does.
Lil feed back to the song, the sub is too much. Try to layer multiple bass sound from sub, maybe add a bit of grittiness of the bass to make it louder so the sub doesn't smacking in your face.
The deeper you get into a session, the more Claude's limitations are shown, compared to OpenAI. Claude doesn't learn during the session. Where as, ChatGPT does have a session based memory.
Coffee makes my mental alertness go from 40 to 80. Who knew? /s Why are you just accepting the Tesla analysis numbers as true? When I actually do the work to check AI answers like that, I find they are mostly wrong. We are doomed if we let AI teach our children.
AI singularity is a hypothetical idea where artificial intelligence is more intelligent than humans. In simpler terms, if machines are smarter than people, a new level of intelligence will be reached that humans can't achieve.
But can it act as my assistant? Can I talk to it, tell it to send an email to another person, and fill in the mail in response to another mail etc? Once AI can work as my personal assistant/worker and interact with people through mails, then we can start talking :))
Did you verify that the graphs and numbers were accurate with the sheet you put in? I'm curious because I did something like the graph one with another au and it did it but the tables and graphs were missing a bunch of stuff or had weird fill ins.
Claude 3.5 has been the first AI that was able to help me figure out some coding problems for my game, and then accurately make changes after I tested the code and requested alterations. I had a really weird edge case that I couldn't find an example for by Google searching, and I didn't understand the underlying system enough to solve it. Claude solved it AND explained the code to me in a way that helped me understand the feature. It's fucking mind blowing.
How are you using it for your code? Just the web browser? I have found a better way (although I haven't tried it yet) with Phind pro and using the Phind extension in VSCode. I need to try it out, just that I don't want to pay $20 per month since I am not actively working on a project at the moment lol.
@@PeterTroutman I tried GPT4 and GPT4o and neither were able to solve my problems effectively. But to be fair, I'm coding in GDscript in the Godot game engine, so maybe Claude just had better access to some data regarding GDscript specifically since it's not a really a mainstream programming language.
Bro calm. I was already blown away by the first example you gave at the beginning of the video and then you kept making it even crazier. This video truly deserves the SHOCKING or STUNNING title
GPTs are cool and APIs can do more. You can also create agents much easier. That said, these results make Claude 3.5 a better base model. I just don't trust Amode. He's gonna want to legislate OS away because of "safety"
@@zerad3307 People have been replaced by these AI's for several years now and it's only going to get worse; most don't hear about it because they aren't directly replacing workers, they just reduce the need for them so to the outsider it seems like the market just got more human competition. Most simple jobs in the digital world can now be replaced if the model is implemented correctly, where you now only need one employee which before took multiple employees, and probably soon more advance programming jobs will also start to feel it. We don't need a fully autonomous AGI to replace most jobs, it only needs to be good enough.
I did a rather complicated coding project with Claude, using my phone, in about 15 minutes that it took me about 2 weeks of 4+ hour coding sessions on my PC to correctly complete. Claude truly is good. Especially since it doesn't break your code randomly by implementing "fixes" to your code behind your back. It is very upfront about explaining what it did and why as well as staying on task.
use the projects feature.. i get alot more requests than 10. the rate limiting isn't ideal but i gotta make my 20 bucks count so just make format your requests to be more thoughtful and provide more output
When I first tried 3.5 Sonnet, I had it work on an XML file. I said "maybe there should be a header element containing X, Y and Z attributes". It did that, AND cleaned up the other places X, Y and Z were defined in the XML. It's such a simple thing, but I've never seen an LLM have the intuition and confidence to do cleanup/optimization without being told.
The insanity of this video is an understatement and I use GPTs all the time to help clean up scripts at work. I knew Claude was different, but this update is simply mind-blowing.
@@mohammadrayyan7851 i have lots of potential purposes for it but a big one was when i tried to use it to review and edit my fiction works... it kept removing literally anything with the slightest violence or eroticism... I'm talking slaps and kisses here, nothing more. It would outright refuse to help most of the time, saying it was against TOS, when it literally wasn't. Wildly overzealous censorship. Useless.
Just helped me refactor a really nasty sql server query with very little context. It actually had me enter a bunch of discovery queries to understand the database. It then provided the most impressive elegant solution to a complicated problem. Very impressive tool. 😊
Its like they're saying "Thanks kiddies, you've let the cat out of the bag, looking at you Open AI, let us take it from here" Open AI are saying "CRAP! Quick bolt on something that looks like Artifacts"
GPT 4o was launched around a month and a half ago, Claude 3.5 is really impressive. I am using both for my work and Claude really does better many times . Both these tools have made my life easy. Open AI probably be preparing to launch another beast of a model in coming weeks to again get 1st position.
I wouldnt bet on it. They've announced to release the voice mode for 4o in autumn and the video mode for early next year. Which places GPT 5 earliest in summer next year, which adds up, seeing how many politicians are now running openAI
I have a suspicion though: Matthew Bermann also tried the snake game and apparently what you got is exactly the same implementation (as I can see visually, code based I am not sure!). What if the AI devs just trained it with hardcoded material for the most common problems that were known that many people would test and many LLMs could not do, so now it just replicated its training data and everyone thinks: "Wow this AI is so good", while it actually is not because the answer is not really freely generated?! Oh and P.S.: I use Claude 3.5 Sonnet via PoE and the coding skills are great, but its "memory" is really bad. After 4 to 5 prompts it completely forget what it implemented so the following code from there on our does not fit the original implementation anymore.
To what extent it is creating the game and to what extent it is simply remembering/sampling a trained code from memory. Internet is filled with SNakelike Mariolike Tetrislike and Doomlike games from CS students all over the world. I agree that it is hard to think up a game outside of the training distribution, but still. (I'm in no way saying that it is bad, I think that it is a grat leap forward I only urge some healthy dose of skepticism).
In my one test I did, it was not impressive.. I tried to get it to make a simple platformer with obstacles that randomly shoot from the ground. It make tons of mistakes like the character can only jump once, gets stuck in the air, the obstacles weren't behaving as requested, etc. and when I asked it to fix these mistakes it would, and then simultaneously create new issues that were arguably even worse.
In ChatGPT I do unique projects. Even big with thousands of lines in different files. It's not just copying, it's remembering, extending, and fixing bugs. But of course, it is not with one prompt. It's like working with Jarvis (without voice)
The type of games was created by people and recreated millions of times. And now artificial pseudo intelligence is able to produce you the garbage and even to edit some shit. It's about combinatorics. Primates: IT'S SINGULARITY! LOL 😅 Clowns 🤡 ARTIFICIAL PSEUDO INTELLIGENCE
I really like your videos. For example I respect you taking your time to properly test the model and not rushing out a video asap. I also like your general tone of talking, which doesn't tend to be overly expressive / hyped. Sort of reminiscent of AI Explained. I've gotten generally quite tired of clickbait and hype-men. I hope you keep it up.
When AI gets this good is the real scary thing. Loss of demand for devs aside, it’s going to become so much easier to have it do the work that we’ll end up having most code be AI generated. That code will become a black box much like an AIs inner logic and whatever bugs it creates along the way could accumulate. All that stuff about it becoming sentient is unrealistic to me but a slow creeping problem of bugs accumulating is probably going to be the real issue.
Not sure why you think ChatGPT couldn't do this. I gave it the same Tetris prompt and it successfully coded the full game with no error messages like you encountered in Claude. I tried the audio visualizer as well and it worked just fine. All with 4o. I'm gonna continue trying in ChatGPT as I continue the video. So far it works just as well if not better that Claude. Would love to hear your thoughts on this. Regardless great video so far. I didn't know you could do all of this cool stuff with AI chat bots!
Thanks to our sponsor Abacus AI. Try their new ChatLLM platform here: chatllm.abacus.ai/?token=aisearch
Prompts used in the video:
create a snake game using python
Create a single HTML page that let's me upload an audio file. Sync that audio with a visualizer like the attached image. Don't use unsupported libraries.
convert this UI design into frontend code
create tetris game using python
create a JS presentation on the health implications of coffee
Create an interactive 2-page infographic on the attached document
Using the neural network diagram I've shared as a reference, please create an interactive HTML/JavaScript animation that demonstrates the flow of data through this network. The animation should include:
1. A visual representation of the network structure, matching the layout in the image.
2. Animated paths showing data flowing from the input layer through the hidden layers to the output layer.
3. The ability to input sample data into the 5 input nodes.
4. Visual feedback showing how the activation of nodes changes based on the input.
5. A simple UI to control the animation speed and reset the simulation.
The animation should be educational, helping users understand how data propagates through a neural network. Please provide the full HTML and JavaScript code needed to run this animation in a web browser.
create an app in one html page that can be used in Artifacts. make an interactive 3D particle cloud, with a max of 100 particles. use Three.js for the simulation. do not use unsupported or 3rd party libraries or functions. create your own functions.
I've been a ChatGPT boy, but your video made me try to experience Anthropic but the UI that I have is quite different to the one that presented in the video, specially it does not have the cool preview option for JS or HTML, is there anything that I am missing ?
@@maoaris What country are you in? Maybe it's not supported yet. Also, did you go here: claude.ai/
Does ChatLLM provide the same Claude 3.5 or they are different?
Don't skip the first 10 seconds. I spent hours trying to get to that stupid slider switch, which, once you've started a conversation, you cannot enable. )(($*$*)_#)_@)_!
Hi
Im just happy anthropic is a valid competitor. Between Google, Apple, and ClosedAi, Anthropic is the protagonist this story needs.
Heel turn incoming
Drunk texting again pal? It’s literally Jeff Bezos.
Apple?
@@zanedurante3709 😂🤣😭 u rite
Data scientist me going unemployed with this one
"when I touch myself I lose the game" well said my friend
LOL😊
and sadly, everyone watching also lost the game!!
first thing I looked for in the comments..
Happens to me too often
What are you, a mormon? 😆
Never ask a man his salary
A crytobro his portfolio,
A woman her age,
An AI fanatic how much power a single ai search consumes
How much?
How much?
@@Stellar-Cowboy2
@@Stellar-Cowboyfive hundred kabillion zillion
@@pet-triflyofficialchannel6528 Something around several hundreds megawatts.
Well, the funny thing about the Snake game is that it created the exact same game for me as well, with the same message and everything. It seems to have been trained specifically on that, or for all we know, it could be identical code pulled from a database.
Probably not a database, just extra, specialized training. It is essentially a learned "skill" now.
interesting, thanks for sharing!
@@theAIsearch I've been wondering whether it's pulling from some database or if it's just incredibly consistent. Remember that other AI guy, Matthew? (I can't recall his name exactly, but it starts with an M.) He does a lot of testing with snake games too. I thought it was an interesting observation.
It seems like the AI might be highly trained on the snake game by now, or it's drawing from a database. It would be interesting to run many different trials and see the exact code that's varying and changing. This could provide insight into what's actually working under the hood. For all we know, it could be utilizing a vast database of open-source code, grabbing that code, and using an AI in the background to modify it.
If they aren't doing that, I'd be disappointed because it would be a great idea.
@@brianlink391 would be interesting to find out exactly for sure
True but you can change anything you want with the game
Sonnet is their middle tier model too, when they release the new Opus it'll be even better.
incredible i feel this is first real progress after gpt4 release
8:36 - Song name : Dumpster Fire...
I noticed in the coffee presentation nothing was said about polyphenols or all-cause-mortality,... Maybe if Claude 3.5 was told to create a graduate level presentation or a presentation to researchers? So far I've been extremely happy with Claude Sonnet.
Absolutely love that it nailed the entire snake game but could not horizontally and vertically center the message when you run into a wall or yourself 😅
lol true
CSS or fronted is hard as f*
There are too many options and can conflict with each other. And AI can't try it out.
With ChatGPT this is the thing that I override. It works best when you use frameworks.
Damn. I imagine myself building my own game just with prompts in the future. So much will change in so many areas.
This would hurt the gaming industry In the future if the ai is keep developing like that
@@hornerfarah2282 By "hurt" you mean indie devs and regular people could create well made games in less time with less experience and it would overall be better for everyone besides AAA studios that milk terrible products every year for money
how horrible
@@ProfShibe To be fair, after all of these AI's getting better and better, I feel that studying programming languages is going to become worthless, it's discouraging me a lot these days
@@darkkittenyt6001I'm literally watching this video as I'm studying computer science, tell me about it
@@darkkittenyt6001 As long as clients are as stupid as they are today, programmers will still be needed. Half of the time they can't explain what they actually want, so they won't get the wanted results from an LLM either. Even if they get a working application out of it, they can't debug issues, scale the codebase or keep track of security flaws. Let them try, they'll come crawling back soon enough.
16:40 “If your job is to create these reports or presentations…”
You’re going to be out of work because they don’t need you to make them anymore.
Somebody needs to provide Anthropic with more compute.
Their rate limit ends pretty soon. 😂😂
yeah the free plan is very limited. i get around 5 messages every 5h
@@theAIsearch Well even the paid version is quite limited if you use it a lot. I mean I'm trying to do a kind of an advanced .php site with it, so I've hit the limit like 3 times today already. If not those limits, then the sky would be the limit lol
@@theAIsearch I get around 4 messages every 3 hours. According to Anthropic if I upgrade, I will get 5x tokens. So, are they saying I'll be stuck at 20 messages per 4 hours or 25 messages every 5 hours in your case? Seems very restrictive.
I pay for the pro version yet I get limited frequently!
@@orthodox_gentleman Good to know. I suspected that may be the case. As someone who literally engages in occasional sessions that are hours long, this does not sound optimal for me.
I just signed up based on this recommendation but my experience was different. Won't generate pictures, work with spreadsheets, etc. Maybe I missed something? " I'm an AI language model designed to analyze and describe images, but I don't have the capability to produce or modify visual content." "I apologize for the confusion, but I cannot directly create, generate, or manipulate any kind of images, including spreadsheet graphs. I'm an AI language model designed to process and generate text-based information."
Have u baught premium?
rewatch the beginning, you have to enable the artifacts feature
Well it can't really create visual, as far as I understand it. Just code. For the many examples he copied the code to another program to run it, and for the others he had it use a web Server.
It can’t make images like chat gpt. ChatGPT uses DALL-E and just gives the prompt to dall e but Claude can’t do that
From now on i will only drink Reine Claude lemonade to be amongst the first to harvest our AGI overlord's ultimate grace.
😂
Have you fact checked the output? Curious if the Tesla summary or other items contain any hallucinations…
I made a 2d fishing game with it. So you click a button called fish and it takes a second and you either collect a clam or fish or crab and you gain money also if you buy anything you get coolness as I added a coolness meter as a joke. Also if you have over 50 dollars you turn yellow and over 100 you turn blue and over 500 turn white and over 1000 neon red. Also a little meter shows when you and what your next rank is. The shop holds 3 upgrades the first is called extra hook the 2nd is extra cash and the 3rd is faster fishing.
Wow holy shit, that's fucking useless.
@FatalityScouter15 redditor attacks 12 year old because he is miserable
I'm glad you're having fun, the game seems cool 😊
Been using them for a while, the ability to get huge context is insane. You can upload the entire great Gatsby book and it can handle it perfectly.
Mind blowing indeed... Thanks for this valuable video
Imagine doing “Creative stuff” with an ai model
Just got it to do a five page interactive document on the timeline of FNAF and it actually helped clear up the story a little more, however it didn't go into any detail regaurding the books as i only specified the video games.
*cries in CS degree tears*
Us bro us end is near 💀
Time to switch to robotics or literally anything else
About time to get another degree!
More than ever ,real computer scientists are needed.
you can always start working on improving the LLMs tbf. They can't do that themselves *yet*
I used to think gpt4 was impressive, but then i discovered Claude 2.0. I haven't looked back since - Claude flat out destroys gpt4 in all aspects i use it for. When 3.5 released i was mind blown by how much it improved further. Can't wait to see the next iteration - if GPT5 isn't going to be x2 better than gpt4, they just won't be able to catch up no matter what they do.
My experience differed wildly, it was very crude. Failed to make, draw, or reproduce a simple visual representation of an electronic schematic when I provided an image.
I have used both and to tell you the truth, i really cant pick just one. GPT is great when it comes to figuring out what to do when problems occur with complicated business logic that involves code from various places, I couldn't get Claude to figure out my problem that easily, its solutions did not work for me. On the other hand, I was facing a very specific problem with local notifications and GPT really could not help at all, like it immediately started repeating the code after some attempts. Claude managed to help me step by step very methodically. I really don't know, maybe it is on me but I really can't choose. I need them both
Can’t we just use this AI tool to make your own personal AI program?
Finally, some good testings. thanks.
The music is a great example of ais limitations. The Leigh man might think that song was amazing but music to those who create is a much more expressive experience. Pop music has become so formulaic that our most simple iterations of ai can recreate top chart songs in an instant or new ones because music has regressed so much the past 20 years. Even if you had bots inside of the most accurate human robot it would never be able to recreate the live version of machine gun by Jimi Hendrix with the same level of emotional projection.
This ai is super cool tho, on the working side of things this ai is ridiculous.
it would be perfect if they fix their censorship issues though i’ve been getting lots of false alarm
Having Patrick star's head explode is a pretty good example of what's going on in my head right now?
Did it really make a video game?
🤯
Where did this come all the sudden?
no, it didn't.
it copied someone's github homework
Now this is excellent! As a no/low code dev, so excited to be entering this revolution🔥
Why don't you use it to fix the audio levels in your UA-cam video , the music's 10 times louder than your voice.
Yes, the audio isn't so great. It even the supposed man talking, sounds very robotic
claude can’t do that yet
That song is SO annoyingly loud
Can’t you give a constructive feedbakc respectfully?
This attitude won’t serve you well in business
I always knew that the ais where purposely lacking due to coporate being corporate because when you really think about it this is just slightly above what ai could actually do at the time
Was waiting for your review on it .
😃
All the LLMs still fail to process image processing. I just made a simple request of adding a hand to an image like holding the given image, and all 3 big LLMs failed (Gemini, Chatgpt-40, and Calude 3.5)
Well I did not expect to see an AI Arms Race in my life, but here we are
exciting times!
Haha I stopped paying openai and went to Claude but the 'projects' tab was over the 'artefacts' tab so I just want to say thank you
Switched my subscription over from GPT to Claude. GPT is pretty good, but I hate how inconsistent its quality can be. One time it's amazing, the other time I use it it's dumb as bricks. Having tried Claude for a bit, it's definitely a step up over GPT in overall quality, at least for writing code for my game.
I used it to integrate a service API it did it in less than 3 minutes, and the service works exactly as intended, The same prompt was used with ChatGPT gave me exceptions and errors
I've never paid for AI but Claude is a masterpiece.
I'm cooked what did I study 4 years for
What was your field of study?
If u studied 4 years to create mediocre boilerplate software, then you were cooked to begin with 💀
Damn, GTA 3 protagonist never needed to speak
I do notice it has allot of limitations too.
Try creating an 'liquid with gravity' using particles inside half a square. that the user is able to rotate and then the particles fall out. its too hard. I get alot of bugs and glitches.
That is mind blowing. 🤯
8:50 I compared the actual spectrogram of the song, and whilst it already has a terrible mixdown and quality, the visualizer just portrays the spectrum entirely wrong. On the far left side, we'd expect the lowest frequency range starting from ~20Hz. The song has not a single signal in that range. The mids and highs are _much_ louder than what this visualizer shows.
Whilst I am impressed of this AI producing valid code, I am not impressed by the result.
This. Not sure it’s “visualizing” but it’s not what we’d assume given what a visualizer is
@@TastySlowCooker it makes me glad that I'm not alone with this sentiment
@@bl8de3 bros 4 lyfe
Destroys this destroys that… and next week there is another update for ChatGPT and all functions are included there too. Really not every little thing should be blown up to „this ends… this destroys… forever….INSANE!!!1!!1“ stop making such titles, because at some point people won’t fall for these superlatives in every title anymore, at least I won’t.
Yeah but it will help to have more competition and therefore more updates
The guy has to eat
this is actually scary how it is able to do this ai has already taken so many graphic design jobs next its going to take game development and programming?
I think this is very cool but we can’t let kids get a hold of this sort of technology at a young age. They are going to lose the ability to think on their own and to critically think
Why do you think all this AI crap is being made so main mainstream, it's to do exactly what you've said
bro they need to think in order to make the prompt anyway
Intellectual property laws will joyfully be under pressure. Power will shift and there will be commercial chaos! The outcome could be ugly or there could be real equalization. The only constant will be change.
Say you use an AI to create a new game based on an idea you prompt for. Then you publish it for sale and it becomes a hit. How do you think the intellectual property ownership should and would be determined? Who owns the judgment?
My view is that IP should have had scale limits and forces may finally see that limits need to be set in IP law. Creativity is a shared existence.
The only problem I have with it is that as long as it doesn’t have some kind of voice mode, I just wont use it as much. It’s still exciting to see the development and progress though.
Hey I dont have the artifacts option, Is there anything specific I have to do to unlock it?
I upload it but it doesn’t work as the vid says
Well shit. I'm sitting here learning how to code and this video pops up. 😭
This is incredible and terrifying.
3.5 opus will be massive
From now on I will use Claude for coding and ChatGPT for advanced math.
I primarily use Chat GPT for image creation. How does Claude compare? I can get fairly consistent and accurate to prompt new characters from Dalle. Most the Art generators don't hone in on a unique art style with a simple prompt the way Dalle does.
Lil feed back to the song, the sub is too much. Try to layer multiple bass sound from sub, maybe add a bit of grittiness of the bass to make it louder so the sub doesn't smacking in your face.
The deeper you get into a session, the more Claude's limitations are shown, compared to OpenAI. Claude doesn't learn during the session. Where as, ChatGPT does have a session based memory.
You touch yourself
You lose the Game
Such great motivating words 👏🏻
Thanks, was wondering boggling where artifacts were hidden - have to log out and log back in to enable, duh.
I don't know how you got it to do that stuff but i couldn't replicate the interactive infographic at all.
Coffee makes my mental alertness go from 40 to 80. Who knew? /s
Why are you just accepting the Tesla analysis numbers as true? When I actually do the work to check AI answers like that, I find they are mostly wrong.
We are doomed if we let AI teach our children.
It won't be long until a.i. will be able to copy paste itself a million times per second and self improve.
Exciting!
what exactly does everybody mean by singularity? how does it look like or what can AI do at this point?
AI singularity is a hypothetical idea where artificial intelligence is more intelligent than humans. In simpler terms, if machines are smarter than people, a new level of intelligence will be reached that humans can't achieve.
It won't even attempt to generate any sort of image for me
But can it act as my assistant? Can I talk to it, tell it to send an email to another person, and fill in the mail in response to another mail etc?
Once AI can work as my personal assistant/worker and interact with people through mails, then we can start talking :))
Yet still gpt-4o has yet to add the features they promised in the “coming weeks” months ago…
lol i wonder when they'll release the voice
it got a ton of backlash so I'm not surprised, we did this lol
Soon llm creators will optimize their code to create snake since all reviewers use it to see if the ai is good. Time to find something new.
Changed the world forever
Did you verify that the graphs and numbers were accurate with the sheet you put in? I'm curious because I did something like the graph one with another au and it did it but the tables and graphs were missing a bunch of stuff or had weird fill ins.
16:40
*if this is your job, you gonna loose it very quickly.
There, I fixed it for you xD
wow live bench is a really good benchmark
I have access to every A.I. and when you prompt them to build on each other, the models hallucinate. How do you get around that?
Claude 3.5 has been the first AI that was able to help me figure out some coding problems for my game, and then accurately make changes after I tested the code and requested alterations.
I had a really weird edge case that I couldn't find an example for by Google searching, and I didn't understand the underlying system enough to solve it.
Claude solved it AND explained the code to me in a way that helped me understand the feature.
It's fucking mind blowing.
But if you didn't have the knowhow you wouldn't know what to ask either. This seems like a good step forward from the idiotic gtps
How are you using it for your code? Just the web browser? I have found a better way (although I haven't tried it yet) with Phind pro and using the Phind extension in VSCode. I need to try it out, just that I don't want to pay $20 per month since I am not actively working on a project at the moment lol.
Did you try GPT4 and it failed? Or just didn't try anything but Claude ?
We need AI to summarize huge amount of data into something usable, consumable by our limited human mind 😂
@@PeterTroutman I tried GPT4 and GPT4o and neither were able to solve my problems effectively. But to be fair, I'm coding in GDscript in the Godot game engine, so maybe Claude just had better access to some data regarding GDscript specifically since it's not a really a mainstream programming language.
Bro calm. I was already blown away by the first example you gave at the beginning of the video and then you kept making it even crazier. This video truly deserves the SHOCKING or STUNNING title
Honestly, if he put shocking or stunning In his titles, I wouldn't even care, this guy always talks about the best.
thanks!
We blew past Shocking and Stunning and straight to speechless
GPTs are cool and APIs can do more. You can also create agents much easier.
That said, these results make Claude 3.5 a better base model. I just don't trust Amode. He's gonna want to legislate OS away because of "safety"
The issue is that those terms are used as clickbait too often, so they have no meaning anymore and only serve to drive sane people away.
Remember when the original ChatGPT came out and it felt revolutionary? That was barely a year and a half ago.
And since then it hasn’t gotten that much better tbh
@@OttoIncandenzayou joking?
@@Anty_Prazahe’s talking about the gpt model. Compared to claude it hasn’t improved that much.
@@OttoIncandenzaI would actually say it’s gotten much worse
we are gonna get so dumb and lazy
So nothing changes.
Literally
"If you're job is to compile these reports imagine how easy..." Yeah, imagine how fired you are.
Your*
You are absolutely wrong about that lol.
@@zerad3307 People have been replaced by these AI's for several years now and it's only going to get worse; most don't hear about it because they aren't directly replacing workers, they just reduce the need for them so to the outsider it seems like the market just got more human competition. Most simple jobs in the digital world can now be replaced if the model is implemented correctly, where you now only need one employee which before took multiple employees, and probably soon more advance programming jobs will also start to feel it. We don't need a fully autonomous AGI to replace most jobs, it only needs to be good enough.
Cry
Scrolled just for this comment
Silence ChatGPT, we have a new king in the real of LLMs.
exciting times! i'm sure claude 3.5 will be dethroned shortly as well
@@theAIsearch if it doesn't get replaced within a month, start using it as the name to beat in video titles.
🗣All hail
@@theAIsearch it can make very convincing chat messages clones look
yeah but free version is limited
I did a rather complicated coding project with Claude, using my phone, in about 15 minutes that it took me about 2 weeks of 4+ hour coding sessions on my PC to correctly complete.
Claude truly is good. Especially since it doesn't break your code randomly by implementing "fixes" to your code behind your back. It is very upfront about explaining what it did and why as well as staying on task.
What language did you use?
@@AaronBlox-h2tenglish
I feel this on a spiritual level
paid 20 bucks, got 10 request limit per 5 hours. No, i will stick to gpt, thanks.
Are you sure? I've been using it for a long time and from my experience the rate limit is more like 50+ requests every few hours
@robbyfrank yes, got 10. I have no idea why, maybe has to do with dynamic allocation
Yikes
Thanks for the warning! Will give it a miss . 10 per hour is useless to me.
use the projects feature.. i get alot more requests than 10. the rate limiting isn't ideal but i gotta make my 20 bucks count so just make format your requests to be more thoughtful and provide more output
Claude 10.1 will code this fun little Elden Ring game with a few simple prompts: Hey Claude, build this please.
😭😭
This is incredible.
The singularity is approaching faster then I thought.
yes!
We best be ready, im sure it's going to be a whole hell of a lot of fun.
That's the point of the singularity actually 😂
@@MrErick1160 It sure is
It stills mimics. It doesn't understand yet, although a lot of jobs can be replaced just mimicking
When I first tried 3.5 Sonnet, I had it work on an XML file. I said "maybe there should be a header element containing X, Y and Z attributes". It did that, AND cleaned up the other places X, Y and Z were defined in the XML. It's such a simple thing, but I've never seen an LLM have the intuition and confidence to do cleanup/optimization without being told.
It's all fun and games until AI starts taking our JERBS.
This is incredible, lost for words, just mind blowing... And it's only getting faster and smarter. Wild!
Too wild!!
The insanity of this video is an understatement and I use GPTs all the time to help clean up scripts at work. I knew Claude was different, but this update is simply mind-blowing.
I disagree. I could have switched to Claude if it wasn't for their absurd censorship. I still use GPT 4o everyday.
What censorship
... OpenAI is certainly not any better about censorship. I find GPT mostly useless for my purposes specifically because of it's censorship.
And what purpose is that? If I may ask ...
@@mohammadrayyan7851 i have lots of potential purposes for it but a big one was when i tried to use it to review and edit my fiction works... it kept removing literally anything with the slightest violence or eroticism... I'm talking slaps and kisses here, nothing more. It would outright refuse to help most of the time, saying it was against TOS, when it literally wasn't. Wildly overzealous censorship. Useless.
Just helped me refactor a really nasty sql server query with very little context. It actually had me enter a bunch of discovery queries to understand the database. It then provided the most impressive elegant solution to a complicated problem. Very impressive tool. 😊
I watch about 5 or 6 youtubers' videos regularly for my AI news but yours has quickly become my favorite.
thank you!
Who do you watch?
@@wezmasta David Shapiro, AI revolution, mattvidpro, Matt wolfe....a few more
@@jaywv1981 can u say few more as well all of them .
Yea diss is de best ai channel
Its like they're saying "Thanks kiddies, you've let the cat out of the bag, looking at you Open AI, let us take it from here" Open AI are saying "CRAP! Quick bolt on something that looks like Artifacts"
My take is OpenAI decided to step out of spotlight, so that when IT apocalypse comes they wont be ones to blame 😅
@denisblack9897 nah, openAI is one of those companies that only cares about shareholder value, AI is built on hype they wouldn't waste that
I asked Chat GPT to create a snake game , Space invaders, and Breakout game… So it must be you. OK it’s not in a window, like that, but it made them
you have to do many prompt to get a fully working game. iI did the same with pong it took me a bit
GPT 4o was launched around a month and a half ago, Claude 3.5 is really impressive. I am using both for my work and Claude really does better many times . Both these tools have made my life easy. Open AI probably be preparing to launch another beast of a model in coming weeks to again get 1st position.
I wouldnt bet on it. They've announced to release the voice mode for 4o in autumn and the video mode for early next year.
Which places GPT 5 earliest in summer next year, which adds up, seeing how many politicians are now running openAI
I have a suspicion though: Matthew Bermann also tried the snake game and apparently what you got is exactly the same implementation (as I can see visually, code based I am not sure!). What if the AI devs just trained it with hardcoded material for the most common problems that were known that many people would test and many LLMs could not do, so now it just replicated its training data and everyone thinks: "Wow this AI is so good", while it actually is not because the answer is not really freely generated?! Oh and P.S.: I use Claude 3.5 Sonnet via PoE and the coding skills are great, but its "memory" is really bad. After 4 to 5 prompts it completely forget what it implemented so the following code from there on our does not fit the original implementation anymore.
To what extent it is creating the game and to what extent it is simply remembering/sampling a trained code from memory. Internet is filled with SNakelike Mariolike Tetrislike and Doomlike games from CS students all over the world. I agree that it is hard to think up a game outside of the training distribution, but still. (I'm in no way saying that it is bad, I think that it is a grat leap forward I only urge some healthy dose of skepticism).
In my one test I did, it was not impressive.. I tried to get it to make a simple platformer with obstacles that randomly shoot from the ground. It make tons of mistakes like the character can only jump once, gets stuck in the air, the obstacles weren't behaving as requested, etc. and when I asked it to fix these mistakes it would, and then simultaneously create new issues that were arguably even worse.
Prompt: « Generate GTA6 with unicorns », Results: Tetris with pink blocks lol
In ChatGPT I do unique projects. Even big with thousands of lines in different files. It's not just copying, it's remembering, extending, and fixing bugs.
But of course, it is not with one prompt. It's like working with Jarvis (without voice)
The type of games was created by people and recreated millions of times. And now artificial pseudo intelligence is able to produce you the garbage and even to edit some shit. It's about combinatorics.
Primates: IT'S SINGULARITY!
LOL 😅 Clowns 🤡
ARTIFICIAL PSEUDO INTELLIGENCE
I really like your videos. For example I respect you taking your time to properly test the model and not rushing out a video asap. I also like your general tone of talking, which doesn't tend to be overly expressive / hyped. Sort of reminiscent of AI Explained. I've gotten generally quite tired of clickbait and hype-men. I hope you keep it up.
Thank you! I appreciate that!
Yes I agree, well done man, keep it sane and calm
When AI gets this good is the real scary thing. Loss of demand for devs aside, it’s going to become so much easier to have it do the work that we’ll end up having most code be AI generated. That code will become a black box much like an AIs inner logic and whatever bugs it creates along the way could accumulate.
All that stuff about it becoming sentient is unrealistic to me but a slow creeping problem of bugs accumulating is probably going to be the real issue.
STOP saying IA is Scary. Sound stupid. This is evolution.
Don't use them if you are so scary.
@@randywa thank you.
@@ПлутаркоФарше you're making no sense my friend
there isn't going to be loss of demand for devs
@@sentakatsuki theres good devs and bad devs. the bad ones might get replaced. the good ones nowhere soon
Artifacts are no longer available in free version, so the entire video didn't age well.
Its good at Javascript, thats about it. Not really mind blowing.
For reference, spent the weekend putting it through its c++ paces and yeah no thanks.
Not sure why you think ChatGPT couldn't do this. I gave it the same Tetris prompt and it successfully coded the full game with no error messages like you encountered in Claude. I tried the audio visualizer as well and it worked just fine. All with 4o. I'm gonna continue trying in ChatGPT as I continue the video. So far it works just as well if not better that Claude. Would love to hear your thoughts on this. Regardless great video so far. I didn't know you could do all of this cool stuff with AI chat bots!