Claude 3.5 Sonnet Projects Tutorial: NEXT LEVEL AI Programming!
Вставка
- Опубліковано 4 лип 2024
- To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/AllAboutAI . You’ll also get 20% off an annual premium subscription.
Claude 3.5 Sonnet Projects Tutorial: NEXT LEVEL AI Programming!
👊 Become a member and get access to GitHub and Code:
/ allaboutai
🤖 Great AI Engineer Course:
scrimba.com/learn/aiengineer?...
🔥 Open GitHub Repos:
github.com/AllAboutAI-YT/easy...
📧 Join the newsletter:
www.allabtai.com/newsletter/
🌐 My website:
www.allabtai.com
Today we use Claude 3.5 Sonnet projects to create a Flappy Bird Boss Battle Game. This is a full tutorial on how i used Claude 3.5 Projects to manage and write the code for the game. The AI Programming feels like next level using Claude 3.5 Sonnet. Check it Out!
00:00 Claude 3.5 Sonnet Projects Intro
02:00 Projects Setup
07:46 Brilliant.org
08:50 Setting up React App
11:08 First Test
12:08 Adding First Sprites and Hitbox
14:36 AI Opponent Setup
19:48 Adding Music and Sounds
24:49 Finishing up the Game
26:51 Playing the Final Game - Наука та технологія
I'm an OpenAI fan, but Claude is really becoming my goto LLM. It's just so much better
Be just an AI fan and it's hard to lose!
I'm just extremely happy that something like Claude 3.5 has come along, the reaction from OpenAi will be crazy, look at what Claude 3.5 can do, the reaction from Openai will be crazy...
What about the lack of internet access and the need to provide your cell number to register?
Not that strange, Claude is the project of the GPT-3 head of research who quit OpenAI and started his own company called Anthropic!
I am not a coder and I'm building a dynamic socioeconomic game with Claude, it's incredible
Cool! love it :)
I followed your approach and adapted it to make it my own.
I created a playable side scroller with controller support too.
I used Udio for background music of level, Elevenlabs for sound effects, Ideogram and GIMP for image assets.
I did this in just few hours too! 🤯
Awsome :) love it! thnx for tuning in!
❤
Nice to see you've unlocked the power of Claude Projects! It's such a productivity multiplier
yeah, looking forward to exploring more :) tnx for tuning in
I was able to make the basics of a traditional roguelike. I am now learning so much using AI as a tutor advisor. Adding features every day so much fun.
love it :)
The real power comes when LLMs like Claude 3.5 Sonnet work together as multiple agent instances. With the agents being connected to tools like IDE, compilers, mid-Journey, etc, the multiple agents can evolve the project development on their own and only need to human to verify their output.
yeah, will be interesting too see where there this goes :)
Yeaaassss ...My thoughts exactly. I can imaging 3 hour would turn into 30 minutes or less. I want to create Fortnite and Counter strike type games.
Coming soon 😅🎉❤
This was great! Awesome workflow. While I may not work the exact same as you, this illustrates how easy it is for folks to jump in.
thnx :) glad you enjoyed it!
This is just insane ! thanks for showing all this
Thanks for sharing 🎉
Claude with projects is pretty impressive.
I like how this project turned out .
thnx :)
Great video, I love the the combo of code and creativity 😀 I have to try this approach for some other games! (In Swift though ;-)
you are connecting all kinds of stuff! i love it!
thnx :)
Another Ai tut masterpiece drops
😮😮😮 sweeet ❗
It's quite nice with artificats.
Really great video. I think this demonstrates to us non-coders how easy it is to create code. Thank you very much.
thnx for tuning in :)
Hei, fikk aldri svar fra deg sist gang, men ville høre om du er kjent med rasa, og om du har erfaring med å legge inn NorBERT large module via huggingface? Jeg vil sette opp min AI salgs bot i vapi med rasa, og bruke NorBERT large model ettersom vapi egentelig ikke støtter norsk, men står fast med å få integrering av custome model.
When you upload to your knowledge base, does that count against your context window?
You nailed it
😂😂 he is so focused while playing.
haha
I’ve never seen you playing so intensely your own created flappy game at the beginning of the video 😂😂😂
haha
Your game reminds me of the game Yie Ar Kung-Fu for the Commodore C64!
I hope that it will soon be possible for everyone to create games/programs for the old box
This is only what well see on youtube for a while now. Oh look this is Ai that did this.
I do like what A.I. is allowing us to do, but I think the real game changer is going to be once A.I. tools are integrated into the tools set like programming languages or game engines and so on, more so if it's a specialised A.I. model designed for the given task, which more likely means a locally run model.
For now, it all feels experimental and a slap on approach, but once integrated, which will only happen once A.I. is a lot more accessible at a local level, that's the game changer.
Fortunately, that's happening, AMD, Intel and ARM's players are setting up the ground works with APU's that have access to a decent amount of A.I. capability, but more importantly, far more memory access, this is important, because if developers are going to integrate A.I. into a lot of core apps that we use, they will only do so once the hardware is there to run it, and are likely only going to do so with locally run models on the hardware as they have more control over it, not to mention security concerns that many will have with online models, especially when working on a project.
I've been playing with the idea of GPU mining rig converted just for this purpose. 🦾
Thanks
Thanks, Awesome ! I really liked how you managed 'Project Knowledge'. For me, it's the most difficult part to handle. It would be very beneficial if you could dedicate a video just to this part to optimize interactions and avoid the chat usage quota.
i dont think there is good solution out there other than calling via api and managing the context directly.
I am waiting for open source port of Claude's artifact functionality where we have control over what is sent to llm. something like promt2ui
@@poisonza Sure, the API is a solution, but I've read some posts saying they managed to get results with just a few prompts, and from what I've seen, the 'Project Knowledge' part has something to do with it. And for the open source, I completely agree!
Yeah this has been my problem. But I didn't even know about this Project side of this. I've just started using Anthropic. I'm like a sponge on it now...lol
It just seems like we could give the api a command list to operate functions. All the same ingredients for a game. It tells you the steps similar each time. With the operation list it would be like a hand shake for agreement. When you make images store them here like this. When you make scripts store them like this. It can come up with idea and as long as it understands how you want to handle the files well everything is auto. With api.
Im starting to think about a “start off” auto game maker with public test function. So basically a community created game. With the fact some human criticism is needed. A platform that can evaluate each section and AI could update according to popular demand.
Well every move could be a poll like voting structure from initial idea list to completion..
Ehh sounds time consuming doesn’t it.
As an example,
The ai could get used to popular function request. Like the complaint about volume. At that point the ai should notice volume levels and automatically intergrate or atleast have it as a top poll choice to vote on. (Volume level slider)
@@playthisnote I'm down for this community building idea❤🦾
Have you tried cursor with Claude tho??
not yet, will tho :) tnx
This video is lit 🎉
thnx man :)
Anthropic take my money! OpenAI the ball is in your court…
ikr.... I wish I was apart of there in house build team. We need to build a squad.
I do love Claude 3.5. but they brag about the affordability of the model, and then are much more strict with use limits than gpt4o which is a bummer... I can't really reliably use this for work unless I buy like 5 accounts
Isnt buying 5 accounts also worth it? I have never coded before and I build an entire custom webshop using Claude wich wouldve cost me thousands of dollars to implement by a dev.
@@RoeJoganShorties Yep, probably in the long run. I think at that point I'd just use the API and pay a few hundred dollars per project and just have the peace of mind that it would get done. These are personal projects, btw - I imagine someone using this for their coding job would get away with having it as an assistant when they need it, and not use it as much as me.
I wonder if AI is able to add juice into the game. Tween animations, particles etc...
Hrm I dont seem to have access to projects on my Claude.
i am very afraid to see this. i am just learning to code. should i continue to learn coding ? i want to be python full stack developer with react as frontend framework
If you use Ai you already are a python full stack developer. React and any other framework.
great:)
It might generate even better variable names if you fix the typo in your system prompt. ;-)
claude reading all community opinion
I was locked out of Claude! Emailed them when it happened 2weeksback and nothing! 🤫 Has anyone else had this problem? How did you resolve?
I want to live in a world where AI is fully developed and very useful for humans. There should be many companies mastering AI so no monopoly
A noob question, is it the free version or is it the paid claude version?
Claude Projects require paid version
I always try to do some crossword, but I never succeed. Graphically yes, but the words never fit together
Try getting this script to work and play your game, challenge: import numpy as np
import cv2
import pyautogui
import time
from collections import deque
import random
import torch
import torch.nn as nn
import torch.optim as optim
# Define the DQN model
class DQN(nn.Module):
def __init__(self, input_shape, n_actions):
super(DQN, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(input_shape[0], 32, kernel_size=8, stride=4),
nn.ReLU(),
nn.Conv2d(32, 64, kernel_size=4, stride=2),
nn.ReLU(),
nn.Conv2d(64, 64, kernel_size=3, stride=1),
nn.ReLU()
)
conv_out_size = self._get_conv_out(input_shape)
self.fc = nn.Sequential(
nn.Linear(conv_out_size, 512),
nn.ReLU(),
nn.Linear(512, n_actions)
)
def _get_conv_out(self, shape):
o = self.conv(torch.zeros(1, *shape))
return int(np.prod(o.size()))
def forward(self, x):
conv_out = self.conv(x).view(x.size()[0], -1)
return self.fc(conv_out)
# Game Environment class
class GameEnvironment:
def __init__(self, screen_region):
self.screen_region = screen_region
def get_screen(self):
screen = np.array(pyautogui.screenshot(region=self.screen_region))
screen = cv2.cvtColor(screen, cv2.COLOR_RGB2GRAY)
screen = cv2.resize(screen, (84, 84))
return np.expand_dims(screen, axis=0)
def take_action(self, action):
# Implement action execution here
# This could involve clicking at specific coordinates or sending keypresses
pass
def get_reward(self):
# Implement reward calculation here
# This could involve analyzing the game state or score
return 0
# Reinforcement Learning Agent
class RLAgent:
def __init__(self, state_shape, n_actions, memory_size=10000, batch_size=32, gamma=0.99, epsilon=1.0, epsilon_decay=0.995, epsilon_min=0.01):
self.state_shape = state_shape
self.n_actions = n_actions
self.memory = deque(maxlen=memory_size)
self.batch_size = batch_size
self.gamma = gamma
self.epsilon = epsilon
self.epsilon_decay = epsilon_decay
self.epsilon_min = epsilon_min
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.model = DQN(state_shape, n_actions).to(self.device)
self.target_model = DQN(state_shape, n_actions).to(self.device)
self.optimizer = optim.Adam(self.model.parameters())
self.criterion = nn.MSELoss()
def remember(self, state, action, reward, next_state, done):
self.memory.append((state, action, reward, next_state, done))
def act(self, state):
if np.random.rand() self.epsilon_min:
self.epsilon *= self.epsilon_decay
def update_target_model(self):
self.target_model.load_state_dict(self.model.state_dict())
# Main training loop
def train_agent(episodes, max_steps):
screen_region = (0, 0, 800, 600) # Adjust this to match your game window
env = GameEnvironment(screen_region)
state_shape = (1, 84, 84)
n_actions = 4 # Adjust this based on the number of possible actions in your game
agent = RLAgent(state_shape, n_actions)
for episode in range(episodes):
state = env.get_screen()
total_reward = 0
for step in range(max_steps):
action = agent.act(state)
env.take_action(action)
next_state = env.get_screen()
reward = env.get_reward()
done = False # Implement a way to determine if the game is over
agent.remember(state, action, reward, next_state, done)
agent.replay()
state = next_state
total_reward += reward
if done:
break
if episode % 10 == 0:
agent.update_target_model()
print(f"Episode: {episode}, Total Reward: {total_reward}, Epsilon: {agent.epsilon}")
# Save the trained model
torch.save(agent.model.state_dict(), "trained_model.pth")
if __name__ == "__main__":
train_agent(episodes=1000, max_steps=500)
If i would be a owner of a big game or movie-studio, i would be concerned. They are simple games now, but the writing on the wall is clear as long as the progress keep going forward. Sure they have their trademarks, and they will try to milk them as much as possible, but otherwize they are probably f****d long term.
i pay for open ai...and lately Claude sonnet has me questioning why I am paying for them... if this is what Anthropic has can't wait to see what open ai does cause they need somethign way better than this or it's time to move on for me.
It's just not enough for me with 200k tokens. I want to make an near infinite repayable text fantasy rpg with classes, crafting, etc in just one prompt but I know it would take tens of millions of tokens lol.
Chat length is horrendous
For real
That's on the free tier though right? I think when you use "projects" as a paid plan it doesn't work in the same way.
The only downside about Claude is not being able to generate images like OpenAI.
Kindalikerightkinda?
Made a couple of videos showing how you can do this same Claude 3.5 Sonnet development but inside of your IDE (Cursor) instead! ua-cam.com/video/x3TRq9N5T-Q/v-deo.html
Awesome video though, I'm glad to see how somebody would do this using Anthropic's Projects feature!
You spent half a day to use Claude to get that level of gameplay, instead of just programming it yourself in what, 4 hours? The game kinda sucks, and primarily using AI means you basically don't know the codebase at all. Even if people suck at programming and would spend a week on programming this, they would learn so much more on spending a week doing that instead of using Claude.
Honestly, I was interested in seeing what "Next level AI Programming" would be, since I thought maybe AI had finally gotten better at it, but ig it still sucks.
man you really need to add a brief channel open transition, your videos always pop right to you talking in a really hard sort of way
Claude is trash. youre paid for scamming people. Claude has a interaction limit. After paying for claude, i asked it 8 question and it said that i could not ask it another question until 4 hours later. All day i prompt chat gemini, and midjourney. 8 questions for Claude. That isnt usable.
Really 🤨
@@tooimpacted Yes: ua-cam.com/video/EghpdIk3pLU/v-deo.html
I wonder if they stealth increased the limit today. I was programming and never saw the 10 message remaining limit. I was annoyed at the limit (and still am) but to call it trash is a bit much. Its superior to Openai in coding. I have both and the difference is noticeable.