I think OpenAI is still holding back something much bigger than the model they just released. In my opinion, they are about a year or more ahead of the competition, as they have only recently reached the performance level of GPT-4. Knowing this, they are not at all concerned about releasing new models or features anytime soon. They launched an "Omni" model to surpass the competition. The competition managed to catch up with that model, and now they've launched a reasoning model. But none of this would scare someone working inside OpenAI on a daily basis, as they were already aware of these advances and knew this would happen. Now, combining: 1. Agents (a base already launched in GPTs), 2. A multimodal model with audio, image, and video, 3. A reasoning model, 4. Long-term memory and planning, All of this together in a single model, which I believe could be a GPT Next, a full Orion, or whatever name they choose to give it. That, indeed, would be something that could surprise someone like Ilya.
@@sushi2721 Ilya Sutskever, OpenAi cofounder and previous head of safety(?). The theory on the internet is that he left (literally the day before 4o was released) the company after failing to oust Sam.
They are, they said it themselves! they want us to get used to the new features slowly and teaching us to use them instead of overwhelming us with everything all at once.
Do you maybe know the best path to translate whole books rn, would chatGPT be better than Gemini for that? i read somwhere that Gemini pro has 1mil token that seems like a larger context window for translating books.
I am a seasoned developer and I can tell you that it’s probably much easier to create small non-complex apps and games without any experience with these models. But when it comes to real world problem solving, I’m still sceptical. Will try it out tonight tho
It can not create complex apps, it will start to halucinate and make errors, and it is very bad at finding its own problems, etc. For example, if you try to ask it to create a simple data editor with a spreadsheet, it will fail. But, if you ask to do it one thing at a time, it can do it, like, instead of asking to write the complete code for the editor at one, ask first to create a window with ribon ui, then add button, then add functionality to them, etc. Not realy meaningful for someone who can program, but for people without programming skilss like me, it can be useful. I need it mainly for some libs in Python to implement GUI's for them, not for commercial purpose etc.
Look. as a fellow dev, i can tell you that if you hold the AI´s hand. it will also hold yours. all my work now is beeing done by AI. but i have to explain and outline the problems and potential solutions in order to prompt it to be effective. AI alone is right now not enough for commercial products, and humans alone nowadays are too slow to keep up. Human + AI = Fast delivery of functional applications
@@ignaciosavi7739 very well said! I’m using AI, and have done since about 2020. I was a beta tester for the GPT (model). It can be very helpful, but right now I’m still doing a much better job than the AI, when it comes to complex parts of the application. AI can do generic things quickly, but you have to be the one sewing it together to something useful.
GPT-4o can help with pre-planning and allow you to create a solid prompt before using the o1 model. This way, you can save some of your weekly query cap. If you have experience as a developer and enjoy learning by doing, you can use AI to handle boring tasks like boilerplate code, time-consuming processes, and problem-solving. Often, it's important to understand how your program works, especially if it's not your solo project... it's hard to put your name on something if you don’t fully grasp how it works. But it's totally fine to experiment with AI while working on your own projects and learn on the way.
I want to thank you David for taking time out of your day to make this informative and educational video. Not many people nowadays share their knowledge with others, but you do!
@@quaziz Most people today explain the basic stuff on YT and then they try to sell you the "complete guide for AI" for a "small price" lol the majority are trying to capitlize the knowledge this dude gives for free!
If you have you're own android I do t see why not. Like physical robot I mean. I'd imagine creating software for the android to learn how to build a home of you're choosing... Now imagine people coming together with android groups... I can go on but you see where I am getting at? Sharing Android robot resources to be self-sufficient?
I'm pretty sure it can help you step by step in building a house by yourself. It's helping me step by step in building my business (business plan, strategy, niche, marketing, social media, guidelines, scripts, angles, planes, timeline, data analytics, payments, web tech stack, landing pages, etc.)
00:04 Build anything with OpenAI 01 in minutes 01:55 OpenAI's 01 model is versatile and powerful 06:12 Build detailed learning plans with OpenAI o1 08:35 Use simple prompts to conserve OpenAI prompts 13:10 Setting up project and installing necessary libraries for visualization 15:38 Customize app to suit needs 21:13 Your only limitation is your imagination and ambition 23:29 Setting up a new item path in OpenAI o1 30:14 Implement real-time updates for inputs in the application. 34:03 Fixing 'Cannot Find Definitions' Error in TypeScript 40:49 Buying a second CH gbt account is a smart investment 45:00 Implementing and modifying code for neural network component 56:24 Utilizing OpenAI's O1 model for building complex projects. 1:00:10 OpenAI o1 created an amazing model. Crafted by Merlin AI.
Hi David, This is a fantastic video! I really enjoyed watching you code and use the AI assistant for help. It’s also quite helpful to see you type prompts in real-time and repeat the statements as you go-teaching us how to prompt effectively. Thanks again, and I’m eager to see more videos like this. Please keep us updated! :)
36:06 we need to have more videos showing the process. Thank you, David, for showing the abs and downs and iterations that this new style of software construction will take. And for those who try to pretend that before this technology that regular programmers did not go through this, you’re talking to a 25 year product manager. Almost all of them went through this. They like to hide it from the outside world.
That’s my favorite part of this. This is like the first time I’m seeing someone honestly put love the actual process you go through to build something you haven’t built before. People also need to understand it’s not a straight forward as others would make you think. So when people say AI would replace programmers by this evening, they need to realize not everyone will want to do all of this to get a final product.
Storng guy strong mindset you pulled through the Analysis Paralysis squeezed through the bottle neck and made it come true That’s Innovation in a nutshell Goodluck bro
i just subscribed, been watching your videos searching ai stuff for ages but just realise every time im looking for useful information on ai you are always the one testing the questions im looking for every time. great stuff dood!!
I was waiting for you to publish new video about o1 ! just applied for the video editor position and hope to get a chance to be tested and showcase my skills. I'll reedit this video as a sample and send it to you soon because I'm excited to watch it 😊😊
By the way, your looks match EXACTLY your content. And that is a compliment. Mr Robot vibes but now with euro-accent. Incredible. Very cool. I will be subscribing sir. I wish I could just download what you know, directly into my brain lol. Cheers.
No, you can't build anything with this yet. I'm sure the day will eventually come, but as a programmer and game developer who has used these LLMs to definitely improve my workflow, I can tell you that we are still far from being able to have AI do all the work. It still gets a lot wrong, and the relatively small context window also limits its ability to comprehend very large complex codes
@@sweetguy2001canada The only real thing that can be said about technology and the future is that it's notoriously hard to predict. Most who have tried have been shown to be wrong. It might happen in a month, or it may take another 100 years.
@@sweetguy2001canada LLM is based on statistics, it takes data and calculates propability distribution, based on it it selects the word, number etc for the output >>>garbage in = garbage out. 1. It is limited to reproducing what already exists, try for example writing code in julia instead of python and see what you will get as an output. 2. It will always make mistakes due to the usage of conditional probabiltiy but not causal inference. Correlation is not equal to causation. So no, not 2 and not even 5 years, we are talking about decades at best. At this moment, they don't do that much to them, they just put in more data, this data increases slightly the precision, because the probabiltiy values for some parts get higher/lower etc, but the point 1 and 2 still remain valid.
1. you can ask for what tools you need 2. you can ask for the ways and hacks you can use in the programs 3. you can ask for directions, and get help 4. you can ask for tutorials and improvement 5. you can ask for feedback! 6. you can give your code to It, and let it analyze it 7. you can ask for code correction! 8. you can ask it to teach you (oh, why doesn't it work?) or (what did I do wrong were?) or just give your code to it and let it debug it sh*t
00:04 Build anything with OpenAI 01 in minutes 01:55 OpenAI's 01 model is versatile and powerful 06:12 Build detailed learning plans with OpenAI o1 08:35 Use simple prompts to conserve OpenAI prompts 13:10 Setting up project and installing necessary libraries for visualization 15:38 Customize app to suit needs 21:13 Your only limitation is your imagination and ambition 23:29 Setting up a new item path in OpenAI o1 30:14 Implement real-time updates for inputs in the application. 34:03 Fixing 'Cannot Find Definitions' Error in TypeScript 40:49 Buying a second CH gbt account is a smart investment 45:00 Implementing and modifying code for neural network component 56:24 Utilizing OpenAI's O1 model for building complex projects. 1:00:10 OpenAI o1 created an amazing model.
The AI advancement is impressive. But this tool is a powerfull one for people who already know about programming and making a good product from the get go. Now, in the future it may be great to turn the studying of computer science into a more serious topic, without the massive webdev influence that is just annoying at this point. I'm a lil scared, but i see a good future for the discipline. And yeah you can learn anything, that's a cool message, but this vid seems a like those 10 hs python courses that promissed you to get a job and you didnt even get to make a calculator and know how it works. You made a thing with an impressive name that's just a graph visualization tool with some libraries. This is not a good example of the capabilities for ai, neither you can sell a product this rough.
Idk man... I am having WAY more luck writing accurate code that implements smoothly using GPT4, 4o writes code whickly but is very lazy. 1o writes code well but does a lot of weird sh*t that isn't consistent with or implements properly with my project. GPT 4 is still the most powerful... I have been using ChatGPT 8+ hours per day since 3.5 was released.
Great content man. @Croci81 has the right path on that. If you build finetune a model for it in the OpenAI-API sandbox, tie it to a java sandbox with your preferred handler (I'm partial to Azure Graph) and the sky is kind of the limit.
You know how to do it? I tried multiple solutions becaude Claude doesnt have Internet connection. But havent found a real proper one except open router who is expensive
Are there any possibility to create program without writing code (like no-code at all)? Like Text-to-program, without the necessity of debug, install libraries etc.? I know there are some no-code frameworks, would be great to connect such tools with chatGpt or something with access to net.
@@renz164 absolutely, that's why it's a tool for professionals, just for a little help, and that's it, it never meant to replace professionals with someone that just knows to use Ai bots, it's an interesting point
I used a chain of thought reasoning layer with GPT4o mini and built the same program with all of the steps. Until I see Orion class models Im gonna be skeptical
Never rely on these AI toys without having the infrastructure component in mind. If your brain get lazy or addicted to AI and a stock market crash happens (And it will), these models will be shut down or put behing absurd paywalls and you are done for, since you have not learned any useful or foundational skills. Prompting is only useful as long as these companies can pay to keep these models running in the cloud.
@@terminator_playz5167 Not very much at all. Metas smallest parameter llama model runs on my pretty low tier gaming pc very well. Obviously, it's limited, but this tech isn't going to just disappear with a stock market crash.
@@YaBoyfelipe The point is mostly being missed around here, but that's fine. What you're basically saying is "Just have enough money to isolate yourself from the crash." Yes, true, get your situation in place, your money AND your relevant SKILLS (technical or non-technical), such that you can adapt to any post-crash environment (implying the SKILL to adapt, a skill has to be exercised. Earning enough to isolate yourself from the world forever is unrealistic in a few short years). Prioritizing AI and AI prerequisites, online or offline, in any way, over foundational skills is beyond ridiculous. Does any of this sound reasonable to you guys? "A crash will happen, but I'm going to delegate all my thinking and adaptability to LLMs, while my brain and biology overall turns to mush, that's fine."? Like, c'mon, do we even need to talk about Data Engineering, SQL Development, Data Analysis/BI, even basic Data Model business understanding and insight extraction, cloud, security, infrastructure, networking and other dozens of examples of areas (which is what most of tech and business really is, not JS or React) where LLMs and AI in general is useless (practically) and bound to be for the next many decades, possible centuries? Have you guys tried asking an LLM to understand 3 different databases from 3 different secured stores (How do you even securely point it down there?), understand everything at the data level, in possibly hundreds of tables with a lot of dirty data, missing data, bogus naming everywhere, then clean up the data (define dirty data first, and don't mess-up), create pipelines, write the SQL (All types) or MDX to pull only the needed data out to a external model, remodel relationships, optimize performance, securely store it, and then answer business questions based on it? A joke for most of the real world. Just one example out of hundreds, but you should get the point, the "shiny" objects are not the real world, and the real world will never ever stop requiring real skills and adaptability from your biological brain, not some LLM, and you need not look far.
just an idea, but could you use chat gpt 4o to write your code knowing it wont work fully and put that code into o1 giving it the context for what its supposed to and ask it to fix the code, so you get the search ability of gpt 4o for the most up to date api's and stuff then basically just use the o1 preview for the actual logic in the code also having a mostly written code to work off from gpt 4o? and by most up to date api's for the code i mean for example for making a trading bot you need a few different apis to work together to make it work which gpt 4o seems to struggle with and as o1 doesn't have the search ability to find out the apis needed this might be a good strategy to get the benefits of both in a simple and easy way
💥 If is that so, why not to build an App to android phones like the Google assistant but run by ChatGPT, using a wake word ? When I say "ok, Talita", ChatGPT voice would wake up, and could execute any funcion I dictate ? Like open programs, close, choose a music, have a conversation, choose a music, take care of my finances... Better than Siri or Alexa. Why this is so hard ? 😮😮😮
@@romanshar5402 Really ? Why nobody did it before, not even OpenAI had this idea like an assistant, as we have on Apple phones run by AI, or as in the movie HER ? 😮
oh ya build a working game of connect 4 in 32- bit MASM assembly language that works for two players. make the board 7X8 and have the game check for wins or draws.
Has anybody here actually tried to build an app this way or similar? I have....ok a MacOS app and sure enough, it didn't work. I had to learn how to program in Swift to make it work.... haha
You took a whole hour and couldn't get to a working poc of a simple 10 minutes project with the help of chatGPT's most advanced ai model. We are not there yet. But it is going to be better soon I hope
Defeats the purpose of learning. Programming is all about solving problems and developing that mindset. This won’t do that for you. Challenge your brain and build it on your own
Why do I have to learn Python if I can just deploy it. You don't have to learn the code You just learn how to deploy it. I just watched this video and now I have a color wheel for my lyrics that looks like something similar to what coffee graders use. I'm literally a coder now. Coders will be like no you're not you don't understand code. If 40 years ago I could deploy my app with code I developed no matter where I got it from everyone would be like wow cool code you got there. And you'll be like no you used a bot. And I will be like I don't care smell my Ferrari dust. Thanks for the piggyback ride.
He didn't learn how to deploy it; he did everything blindly and simply ran the local server. Enterprise applications demand much more, such as server configuration, microservice connections, distributed systems, encryption, databases, orchestration, clusters, testing pipelines, monitoring, performance optimization and more. And this all is what my company requires from me apart for scripts writing. There will be demand on skilled developers for a long time. Keep learning, and you'll stay ahead. Mastery takes time and effort, and that’s what will set you apart in the long run. You may not do exactly the same thing like now (writing a code) but the critical thinking, problem solving and ability to be divers and adaptive will be in high demand.
I just realized how unfair TH-cam is. 44k views on a video where the creator doesn’t know anything about programming, and my channel Nomadic Geek doesn’t even have that amount on all videos all together 😢
Also a mistake I see a LOT: In the video you say, “don’t ask simple questions” why? GPT4 & o1 is the best for approaching a topic with nuance. ---THAT BEING SAID--- you do not need a word salad to get what you need. Be clear, be concise, be logical. I see a lot of people making a prompt a page long and it’s like WHY
I am very disappointed. Since the release of o1, the results 4o and o1 gave were f*cking terrible. It refuses to listen, before the release, I could input a very large prompt with clear instructions and it executed it almost perfectly. With a few more prompts necessary to have it create exactly what I asked for. But after the release, I have to tell it the same thing 10 times before it will do it, and that's with very small instructions, and when I continue with the next steps of my project, it just deletes what we already had and it becomes an infinite loop of mistakes...
🔥 Start making money with AI: www.skool.com/new-society
DAVID CAN YOU PLEASE SHARE YOUR VIEW ON REPLIT AGENT PLEASE
##
to prove it im a kid out of highschool but no job source now we can. make triple a games ! HAH
import pygame
import random
import socket
import threading
import pickle
import miniupnpc
# Initialize Pygame
pygame.init()
# Screen dimensions
WIDTH, HEIGHT = 800, 600
screen = pygame.display.set_mode((WIDTH, HEIGHT))
pygame.display.set_caption("Club Pokémon")
# Define colors
BLACK = (0, 0, 0)
LIGHT_GREEN = (0, 255, 0)
WHITE = (255, 255, 255)
GRAY = (150, 150, 150)
# Tile size
TILE_SIZE = 64
# Sample tile map
tile_map = [
[0, 0, 0, 0, 0, 0, 0, 0, 1, 1],
[0, 0, 0, 0, 2, 2, 2, 0, 0, 0],
[0, 3, 3, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 2, 2, 2, 0, 0],
[0, 0, 0, 4, 0, 0, 0, 0, 0, 0],
]
# Player class
class Player:
def __init__(self, x, y):
self.x = x
self.y = y
def move(self, dx, dy):
self.x += dx
self.y += dy
def serialize(self):
return (self.x, self.y)
@staticmethod
def deserialize(data):
x, y = data
return Player(x, y)
# Networking setup
players = {}
def server_thread():
server_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_socket.bind(('0.0.0.0', 12345)) # Bind to all interfaces
server_socket.listen(5)
while True:
client_socket, addr = server_socket.accept()
print(f"Connection from {addr}")
players[addr] = Player(1 * TILE_SIZE, 1 * TILE_SIZE)
threading.Thread(target=client_handler, args=(client_socket, addr)).start()
def client_handler(client_socket, addr):
while True:
try:
data = client_socket.recv(1024)
if not data:
break
# Update player position based on received data
players[addr] = Player.deserialize(pickle.loads(data))
# Broadcast player positions to all clients
positions = {addr: player.serialize() for addr, player in players.items()}
for player_addr in players.keys():
player_socket = players[player_addr].client_socket
player_socket.send(pickle.dumps(positions))
except Exception as e:
print(f"Connection error: {e}")
break
client_socket.close()
del players[addr]
def draw_overworld():
# Draw the overworld
for row_index, row in enumerate(tile_map):
for col_index, tile in enumerate(row):
color = WHITE if tile == 0 else GRAY
pygame.draw.rect(screen, color, (col_index * TILE_SIZE, row_index * TILE_SIZE, TILE_SIZE, TILE_SIZE))
def draw_menu():
screen.fill(BLACK)
font = pygame.font.Font(None, 48)
title = font.render("Club Pokémon", True, WHITE)
subtitle = font.render("Press H to Host or J to Join a Server", True, WHITE)
screen.blit(title, (WIDTH // 2 - title.get_width() // 2, HEIGHT // 2 - 50))
screen.blit(subtitle, (WIDTH // 2 - subtitle.get_width() // 2, HEIGHT // 2 + 10))
# Host button
host_button = pygame.Rect(WIDTH // 2 - 100, HEIGHT // 2 + 60, 200, 50)
pygame.draw.rect(screen, LIGHT_GREEN, host_button)
button_text = font.render("Host Server", True, BLACK)
screen.blit(button_text, (host_button.x + (host_button.width - button_text.get_width()) // 2,
host_button.y + (host_button.height - button_text.get_height()) // 2))
return host_button
def automatic_port_forward():
try:
upnp = miniupnpc.UPnP()
upnp.discoverdelay = 200
upnp.discover()
upnp.selectigd()
external_ip = upnp.externalipaddress()
print(f"External IP: {external_ip}")
port = 12345
upnp.addportmapping(port, 'TCP', upnp.lanaddr, port, 'Club Pokémon', '', 0)
print(f"Port {port} forwarded")
except Exception as e:
print(f"Port forwarding failed: {e}")
def main():
running = True
server_running = False
player = Player(1 * TILE_SIZE, 1 * TILE_SIZE)
while running:
host_button = draw_menu() # Draw the main menu
for event in pygame.event.get():
if event.type == pygame.QUIT:
running = False
if event.type == pygame.MOUSEBUTTONDOWN:
if event.button == 1: # Left mouse button
if host_button.collidepoint(event.pos) and not server_running:
server_running = True
threading.Thread(target=server_thread, daemon=True).start()
print("Hosting server...")
automatic_port_forward() # Attempt port forwarding
if event.type == pygame.KEYDOWN:
if event.key == pygame.K_j:
print("Joining server...") # Placeholder for joining logic
if server_running:
draw_overworld() # Draw the overworld
pygame.draw.rect(screen, LIGHT_GREEN, (player.x, player.y, TILE_SIZE, TILE_SIZE))
# Draw other players
for addr, p in players.items():
if addr != 'localhost':
pygame.draw.rect(screen, WHITE, (p.x, p.y, TILE_SIZE, TILE_SIZE))
pygame.display.flip()
pygame.quit()
if __name__ == "__main__":
main()
⭐⭐⭐⭐⭐: 2025 📈
I think OpenAI is still holding back something much bigger than the model they just released. In my opinion, they are about a year or more ahead of the competition, as they have only recently reached the performance level of GPT-4. Knowing this, they are not at all concerned about releasing new models or features anytime soon. They launched an "Omni" model to surpass the competition. The competition managed to catch up with that model, and now they've launched a reasoning model. But none of this would scare someone working inside OpenAI on a daily basis, as they were already aware of these advances and knew this would happen.
Now, combining:
1. Agents (a base already launched in GPTs),
2. A multimodal model with audio, image, and video,
3. A reasoning model,
4. Long-term memory and planning,
All of this together in a single model, which I believe could be a GPT Next, a full Orion, or whatever name they choose to give it. That, indeed, would be something that could surprise someone like Ilya.
whos llya?
@@sushi2721 Ilya Sutskever, OpenAi cofounder and previous head of safety(?). The theory on the internet is that he left (literally the day before 4o was released) the company after failing to oust Sam.
They are, they said it themselves! they want us to get used to the new features slowly and teaching us to use them instead of overwhelming us with everything all at once.
Do you maybe know the best path to translate whole books rn, would chatGPT be better than Gemini for that? i read somwhere that Gemini pro has 1mil token that seems like a larger context window for translating books.
Yeah gpt 5
I am a seasoned developer and I can tell you that it’s probably much easier to create small non-complex apps and games without any experience with these models. But when it comes to real world problem solving, I’m still sceptical. Will try it out tonight tho
It can not create complex apps, it will start to halucinate and make errors, and it is very bad at finding its own problems, etc.
For example, if you try to ask it to create a simple data editor with a spreadsheet, it will fail. But, if you ask to do it one thing at a time, it can do it, like, instead of asking to write the complete code for the editor at one, ask first to create a window with ribon ui, then add button, then add functionality to them, etc.
Not realy meaningful for someone who can program, but for people without programming skilss like me, it can be useful. I need it mainly for some libs in Python to implement GUI's for them, not for commercial purpose etc.
@@nikolaizaicev9297Just like The old models then 😊
Absolutely , not even a functional website...bugs bugs@@nikolaizaicev9297
Look. as a fellow dev, i can tell you that if you hold the AI´s hand. it will also hold yours. all my work now is beeing done by AI. but i have to explain and outline the problems and potential solutions in order to prompt it to be effective. AI alone is right now not enough for commercial products, and humans alone nowadays are too slow to keep up. Human + AI = Fast delivery of functional applications
@@ignaciosavi7739 very well said! I’m using AI, and have done since about 2020. I was a beta tester for the GPT (model). It can be very helpful, but right now I’m still doing a much better job than the AI, when it comes to complex parts of the application. AI can do generic things quickly, but you have to be the one sewing it together to something useful.
GPT-4o can help with pre-planning and allow you to create a solid prompt before using the o1 model. This way, you can save some of your weekly query cap.
If you have experience as a developer and enjoy learning by doing, you can use AI to handle boring tasks like boilerplate code, time-consuming processes, and problem-solving. Often, it's important to understand how your program works, especially if it's not your solo project... it's hard to put your name on something if you don’t fully grasp how it works. But it's totally fine to experiment with AI while working on your own projects and learn on the way.
Dont tell me what to do
o1 API is only available to tier 4 and above I believe
hey btw its no longer weekly, its daily
@@scriptod91 15 and 50 prompts per day?
Good comment man. All these guys are so delusional to think they can finish these Upwork tasks with just AI
I want to thank you David for taking time out of your day to make this informative and educational video. Not many people nowadays share their knowledge with others, but you do!
Isn’t that what TH-cam is for?
@@quaziz sadly 90% is complete bullshit content
@@quaziz Most people today explain the basic stuff on YT and then they try to sell you the "complete guide for AI" for a "small price" lol the majority are trying to capitlize the knowledge this dude gives for free!
Can you build a house ?
If you have you're own android I do t see why not. Like physical robot I mean. I'd imagine creating software for the android to learn how to build a home of you're choosing... Now imagine people coming together with android groups... I can go on but you see where I am getting at? Sharing Android robot resources to be self-sufficient?
Sure you can. Just ask o1 how to build a house 😅
I'm pretty sure it can help you step by step in building a house by yourself. It's helping me step by step in building my business (business plan, strategy, niche, marketing, social media, guidelines, scripts, angles, planes, timeline, data analytics, payments, web tech stack, landing pages, etc.)
Can you buy a pizza with Bitcoin?
😂😂
00:04 Build anything with OpenAI 01 in minutes
01:55 OpenAI's 01 model is versatile and powerful
06:12 Build detailed learning plans with OpenAI o1
08:35 Use simple prompts to conserve OpenAI prompts
13:10 Setting up project and installing necessary libraries for visualization
15:38 Customize app to suit needs
21:13 Your only limitation is your imagination and ambition
23:29 Setting up a new item path in OpenAI o1
30:14 Implement real-time updates for inputs in the application.
34:03 Fixing 'Cannot Find Definitions' Error in TypeScript
40:49 Buying a second CH gbt account is a smart investment
45:00 Implementing and modifying code for neural network component
56:24 Utilizing OpenAI's O1 model for building complex projects.
1:00:10 OpenAI o1 created an amazing model.
Crafted by Merlin AI.
What are u trying to say ?
@@Mosigma-g1y What are u trying to say ?
So video is about basic promting?
Thank you goat
You are the most excited and on-top-of-news guy I have come across ! Amazing content and solid work ! Keep it up brother !
bro is always cooked! the news cant keep up to him aha
A combination of vercel v0(front end and design) and claude worked really well for me!
Hi David,
This is a fantastic video! I really enjoyed watching you code and use the AI assistant for help. It’s also quite helpful to see you type prompts in real-time and repeat the statements as you go-teaching us how to prompt effectively.
Thanks again, and I’m eager to see more videos like this. Please keep us updated! :)
this has to be ai
36:06 we need to have more videos showing the process.
Thank you, David, for showing the abs and downs and iterations that this new style of software construction will take. And for those who try to pretend that before this technology that regular programmers did not go through this, you’re talking to a 25 year product manager.
Almost all of them went through this. They like to hide it from the outside world.
That’s my favorite part of this. This is like the first time I’m seeing someone honestly put love the actual process you go through to build something you haven’t built before. People also need to understand it’s not a straight forward as others would make you think. So when people say AI would replace programmers by this evening, they need to realize not everyone will want to do all of this to get a final product.
@@zubbyemodi6356 Fully agree. Helps refine my process.
Wow this is literally the future love your channel
Storng guy strong mindset you pulled through the Analysis Paralysis squeezed through the bottle neck and made it come true That’s Innovation in a nutshell
Goodluck bro
i just subscribed, been watching your videos searching ai stuff for ages but just realise every time im looking for useful information on ai you are always the one testing the questions im looking for every time. great stuff dood!!
I was waiting for you to publish new video about o1 !
just applied for the video editor position and hope to get a chance to be tested and showcase my skills. I'll reedit this video as a sample and send it to you soon because I'm excited to watch it 😊😊
Bro just decided to choose the most diffcult sort of problem to demonstrate O1 .. I like the confidence
By the way, your looks match EXACTLY your content. And that is a compliment. Mr Robot vibes but now with euro-accent. Incredible. Very cool. I will be subscribing sir. I wish I could just download what you know, directly into my brain lol. Cheers.
No, you can't build anything with this yet. I'm sure the day will eventually come, but as a programmer and game developer who has used these LLMs to definitely improve my workflow, I can tell you that we are still far from being able to have AI do all the work. It still gets a lot wrong, and the relatively small context window also limits its ability to comprehend very large complex codes
2 years max
@@sweetguy2001canada The only real thing that can be said about technology and the future is that it's notoriously hard to predict. Most who have tried have been shown to be wrong. It might happen in a month, or it may take another 100 years.
@@sweetguy2001canada
LLM is based on statistics, it takes data and calculates propability distribution, based on it it selects the word, number etc for the output >>>garbage in = garbage out.
1. It is limited to reproducing what already exists, try for example writing code in julia instead of python and see what you will get as an output.
2. It will always make mistakes due to the usage of conditional probabiltiy but not causal inference. Correlation is not equal to causation.
So no, not 2 and not even 5 years, we are talking about decades at best.
At this moment, they don't do that much to them, they just put in more data, this data increases slightly the precision, because the probabiltiy values for some parts get higher/lower etc, but the point 1 and 2 still remain valid.
@@sweetguy2001canada modern llms tech is probably super close to a plateau or already reached it. What my intuition tells me.
@@tomashgrey2211o1 literally broke that plateau
Your worth ethic is insane
1. you can ask for what tools you need
2. you can ask for the ways and hacks you can use in the programs
3. you can ask for directions, and get help
4. you can ask for tutorials and improvement
5. you can ask for feedback!
6. you can give your code to It, and let it analyze it
7. you can ask for code correction!
8. you can ask it to teach you (oh, why doesn't it work?) or (what did I do wrong were?) or just give your code to it and let it debug it
sh*t
00:04 Build anything with OpenAI 01 in minutes
01:55 OpenAI's 01 model is versatile and powerful
06:12 Build detailed learning plans with OpenAI o1
08:35 Use simple prompts to conserve OpenAI prompts
13:10 Setting up project and installing necessary libraries for visualization
15:38 Customize app to suit needs
21:13 Your only limitation is your imagination and ambition
23:29 Setting up a new item path in OpenAI o1
30:14 Implement real-time updates for inputs in the application.
34:03 Fixing 'Cannot Find Definitions' Error in TypeScript
40:49 Buying a second CH gbt account is a smart investment
45:00 Implementing and modifying code for neural network component
56:24 Utilizing OpenAI's O1 model for building complex projects.
1:00:10 OpenAI o1 created an amazing model.
Thanks! This helps a ton!
The AI advancement is impressive. But this tool is a powerfull one for people who already know about programming and making a good product from the get go. Now, in the future it may be great to turn the studying of computer science into a more serious topic, without the massive webdev influence that is just annoying at this point. I'm a lil scared, but i see a good future for the discipline.
And yeah you can learn anything, that's a cool message, but this vid seems a like those 10 hs python courses that promissed you to get a job and you didnt even get to make a calculator and know how it works. You made a thing with an impressive name that's just a graph visualization tool with some libraries. This is not a good example of the capabilities for ai, neither you can sell a product this rough.
I love your mindset, inspiring
Idk man... I am having WAY more luck writing accurate code that implements smoothly using GPT4, 4o writes code whickly but is very lazy. 1o writes code well but does a lot of weird sh*t that isn't consistent with or implements properly with my project.
GPT 4 is still the most powerful... I have been using ChatGPT 8+ hours per day since 3.5 was released.
Great content man. @Croci81 has the right path on that. If you build finetune a model for it in the OpenAI-API sandbox, tie it to a java sandbox with your preferred handler (I'm partial to Azure Graph) and the sky is kind of the limit.
I wonder when it’ll be out of preview and to expect the full models & less restrictive usage limits. 30 messages a week isn’t that much.
Claude combined with 4o1/4o1 mini can be a powerful combination here.
You know how to do it?
I tried multiple solutions becaude Claude doesnt have Internet connection.
But havent found a real proper one except open router who is expensive
This reminds me of when I was studying python, it was song, now we can do this? Thanks for sharing, I don’t know what I will do next.
Are there any possibility to create program without writing code (like no-code at all)? Like Text-to-program, without the necessity of debug, install libraries etc.? I know there are some no-code frameworks, would be great to connect such tools with chatGpt or something with access to net.
not yet but there are definitely working on it
Try searching for low code / no code platforms. They’re just a Google away.
wow, now imagine a world that you can build absolutely anything, with zero experience, and that world is the one we're living in!
This will create a generation of bad programmers and applications with security vulnerabilities.
@@renz164 absolutely, that's why it's a tool for professionals, just for a little help, and that's it, it never meant to replace professionals with someone that just knows to use Ai bots, it's an interesting point
I used a chain of thought reasoning layer with GPT4o mini and built the same program with all of the steps.
Until I see Orion class models Im
gonna be skeptical
This video seems profound. Thank you for sharing.
Never rely on these AI toys without having the infrastructure component in mind. If your brain get lazy or addicted to AI and a stock market crash happens (And it will), these models will be shut down or put behing absurd paywalls and you are done for, since you have not learned any useful or foundational skills. Prompting is only useful as long as these companies can pay to keep these models running in the cloud.
@@BlackByDopeDemand and do you realize how much more computing power a person needs to run these LLM's offline?
@@terminator_playz5167 Not very much at all. Metas smallest parameter llama model runs on my pretty low tier gaming pc very well. Obviously, it's limited, but this tech isn't going to just disappear with a stock market crash.
I have a mid tier GPU and the offline models run fine. It’s actually impressive.
Then the real problem to me seems to be money, get your money up, so when a market crash happens, you can solve any problem that arises
@@YaBoyfelipe The point is mostly being missed around here, but that's fine. What you're basically saying is "Just have enough money to isolate yourself from the crash." Yes, true, get your situation in place, your money AND your relevant SKILLS (technical or non-technical), such that you can adapt to any post-crash environment (implying the SKILL to adapt, a skill has to be exercised. Earning enough to isolate yourself from the world forever is unrealistic in a few short years).
Prioritizing AI and AI prerequisites, online or offline, in any way, over foundational skills is beyond ridiculous. Does any of this sound reasonable to you guys? "A crash will happen, but I'm going to delegate all my thinking and adaptability to LLMs, while my brain and biology overall turns to mush, that's fine."? Like, c'mon, do we even need to talk about Data Engineering, SQL Development, Data Analysis/BI, even basic Data Model business understanding and insight extraction, cloud, security, infrastructure, networking and other dozens of examples of areas (which is what most of tech and business really is, not JS or React) where LLMs and AI in general is useless (practically) and bound to be for the next many decades, possible centuries? Have you guys tried asking an LLM to understand 3 different databases from 3 different secured stores (How do you even securely point it down there?), understand everything at the data level, in possibly hundreds of tables with a lot of dirty data, missing data, bogus naming everywhere, then clean up the data (define dirty data first, and don't mess-up), create pipelines, write the SQL (All types) or MDX to pull only the needed data out to a external model, remodel relationships, optimize performance, securely store it, and then answer business questions based on it? A joke for most of the real world. Just one example out of hundreds, but you should get the point, the "shiny" objects are not the real world, and the real world will never ever stop requiring real skills and adaptability from your biological brain, not some LLM, and you need not look far.
Yo man, thanks a lot for the video. Keep it up champ, love ur content in general. Btw, u think o1 can work with web3 dev/programing ?
Still not having programing knowledge, it is quite hard. One error and your code is broken.
Why not just have o1 give direct commands to the cursor editor
How would you do this though?
Kindly advise
You Only get what 30 prompts before you run out and have to wait a whole week before you get access again..
Seems like he did more than 30 prompts though?
@@zubbyemodi6356 There was only 3 o1 prompts, he utilized other GPT's for smaller scale basic prompts to limit wasting prompt usage on o1.
It's 50 prompts a week for o1 preview & 50 prompts a day for o1 mini now.
@@undefined428 oh awesome. Thanks!
Thank You! Great Example man appreciate you! 🥳
just an idea, but could you use chat gpt 4o to write your code knowing it wont work fully and put that code into o1 giving it the context for what its supposed to and ask it to fix the code, so you get the search ability of gpt 4o for the most up to date api's and stuff then basically just use the o1 preview for the actual logic in the code also having a mostly written code to work off from gpt 4o?
and by most up to date api's for the code i mean for example for making a trading bot you need a few different apis to work together to make it work which gpt 4o seems to struggle with and as o1 doesn't have the search ability to find out the apis needed this might be a good strategy to get the benefits of both in a simple and easy way
Nice strategy
💥 If is that so, why not to build an App to android phones like the Google assistant but run by ChatGPT, using a wake word ? When I say "ok, Talita", ChatGPT voice would wake up, and could execute any funcion I dictate ? Like open programs, close, choose a music, have a conversation, choose a music, take care of my finances... Better than Siri or Alexa. Why this is so hard ? 😮😮😮
You can really do that
@@romanshar5402 Really ? Why nobody did it before, not even OpenAI had this idea like an assistant, as we have on Apple phones run by AI, or as in the movie HER ? 😮
iOS will have ChatGPT integrated with Siri soon and Alexa is being integrated with Claude as well, just gotta wait.
It’s not that easy even with all the ai tools we have. Complex tasks require very precise prompts into the ai to function exactly as intended.
Excellent Video Man, thank you! Subscribed.
oh ya build a working game of connect 4 in 32- bit MASM assembly language that works for two players. make the board 7X8 and have the game check for wins or draws.
Has anybody here actually tried to build an app this way or similar? I have....ok a MacOS app and sure enough, it didn't work. I had to learn how to program in Swift to make it work.... haha
Look at the good part at least you learn swift ha ha😂
I built a full-stack react native app (google firebase backend) that works on ios and android using gpt4.
Someone should use o1 to build or improve an open source free version of curser, that would be a good use
You can do similar with VS Code and ClaudeDev (OpenRouter o1 integrated)
tell it to thinnk for longer to really juice it
Could you share the link to this conversation you had with o1? I would love to have the learning plan it created for you
I still believe in Meta's open source AI Idea.
Why don't you use OpenRouter in your video ?
Hi David. We are launching a platform in Japan and wondered how we might connect?
Super video Ondrej
Chatgpt 4-O is the most advanced version
Bro I must ask, what did you grab 3.75g more or less? and how far into it at the point of making this video? :P
Can you build your old channel agean?
You took a whole hour and couldn't get to a working poc of a simple 10 minutes project with the help of chatGPT's most advanced ai model. We are not there yet.
But it is going to be better soon I hope
Does this mean we could use the OpenRouter chat but does it mean that we could use the o1 mini API?
Great video thanks for sharing
Defeats the purpose of learning. Programming is all about solving problems and developing that mindset. This won’t do that for you. Challenge your brain and build it on your own
Just use Replit. Why the complication? Just to make content?
How can we do that with replit
Can cursor use open router directly?
Thank you!
I need a bridge built, what's the prompt to achieve this?😅
you're gonna have to prompt o1 to abolish republics and replace them with radical democracies for that one
Can u build a money printer?
What is the AI chat extension David uses on VScode?
It's called "Cursor" and it's not a VS Code extension but rather it's own Editor
Bro ty jsy zabijak hledal jsem jak zacit s GPTo1 keep the AI Grind going
Why do I have to learn Python if I can just deploy it. You don't have to learn the code You just learn how to deploy it. I just watched this video and now I have a color wheel for my lyrics that looks like something similar to what coffee graders use.
I'm literally a coder now. Coders will be like no you're not you don't understand code.
If 40 years ago I could deploy my app with code I developed no matter where I got it from everyone would be like wow cool code you got there.
And you'll be like no you used a bot. And I will be like I don't care smell my Ferrari dust. Thanks for the piggyback ride.
He didn't learn how to deploy it; he did everything blindly and simply ran the local server. Enterprise applications demand much more, such as server configuration, microservice connections, distributed systems, encryption, databases, orchestration, clusters, testing pipelines, monitoring, performance optimization and more. And this all is what my company requires from me apart for scripts writing.
There will be demand on skilled developers for a long time. Keep learning, and you'll stay ahead. Mastery takes time and effort, and that’s what will set you apart in the long run.
You may not do exactly the same thing like now (writing a code) but the critical thinking, problem solving and ability to be divers and adaptive will be in high demand.
Are OpenSource projects using AI to enhance?
But you cannot upload data right?
my opinion its that it costs more, takes more time and output the same bullshit of gpt4, sonnet 3.5 still king
Crazy AF!
This guy broke the system, he prompted o1 for a money method and it created a website for him to sell unlimited uses of itself😂😂😂😂
Great, thanks 👍
I just realized how unfair TH-cam is. 44k views on a video where the creator doesn’t know anything about programming, and my channel Nomadic Geek doesn’t even have that amount on all videos all together 😢
skill issue
Sure, I guess. I think it has to do with the fact that I’m using an ai voice in my tutorials
Maybe you should stop complaining and start building your channel. Just saying.
@@vladonutueu nah. Dont care anylonger. As I said, just realized.
@@vladonutueu just to clarify, I do not begrudge the creator his success. He does this better than me obviously and he should be rewarded.
37:00 - You can build anything !
Can you build a snowman ?
Crazy russian/czech accent.
Teach people to use aider and Claude Dev
what extension is he using in VS code 24:05
Excellent.
Looks like AI soon will replace programmers...
Where is the link to o1
very good
I asked o1mini if it was GPT5 and it responded that it’a architecture was based on GPT4.
Fix my oven door
Can it build my credit?
Can I build a 30y/o blonde baddie with max-level gyatt ? I swear it's for educational purposes only... 😏
🔥🔥🔥
Can it build a car, or a house? No.
Be the first to build Doom with it!
Build a new planet?
I'm also new with cursor! Lol
Also a mistake I see a LOT:
In the video you say, “don’t ask simple questions” why? GPT4 & o1 is the best for approaching a topic with nuance.
---THAT BEING SAID--- you do not need a word salad to get what you need. Be clear, be concise, be logical. I see a lot of people making a prompt a page long and it’s like WHY
Thanks
Eh I asked it to answer a simple question and it was like idk lol a bunch of times
I am very disappointed. Since the release of o1, the results 4o and o1 gave were f*cking terrible.
It refuses to listen, before the release, I could input a very large prompt with clear instructions and it executed it almost perfectly. With a few more prompts necessary to have it create exactly what I asked for.
But after the release, I have to tell it the same thing 10 times before it will do it, and that's with very small instructions, and when I continue with the next steps of my project, it just deletes what we already had and it becomes an infinite loop of mistakes...
W yall
Anything you say
Based on his video, I can conclude that claude is better than this on programming 😅
Cenim cenim
Build me a car