AI Agents are changing the world, let's build one from scratch
Вставка
- Опубліковано 1 жов 2024
- Do you want to join my team? Apply here: forms.gle/2iz4...
If you're serious about AI, and want to learn how to build Agents, join my community: www.skool.com/...
Follow me on Twitter - x.com/DavidOnd...
Please Subscribe.
In this video you'll learn how to code your own AI Agent from scratch, without using any Agent frameworks.
🔥 If you're serious about AI, and want to learn how to build Agents, join my community: www.skool.com/new-society
For the past year I watched about 300 videos on AI agents (all of your videos + a lot of other UA-camrs) I've yet to find any real life business use-case of AI agents , besides simple basic research and summary on a topic/idea or coding, where 2-3 AI "agents" role play with each other. That's about it.
I own a business with 17 people in the eCommerce niche with a lot of automation possibilities but I haven't really seen anything even remotely practical and useful (not only for my business but for almost any business)
Everything is "in the future" but not right now.
@@rodpamora Yea, I can imagine is useful for consulting or similiar kind of work, but as for now I can't really find even one business, that use it for something other than that.
I really hope soon that will change, because I can't wait to automate some stuff.
So far the most useful thing for my business is Stable Diffusion with custom trained checkpoints.
It literally saved me thousands of dollars per year for product photography
You probably should talk to David honestly
@@HakaiKaien Naah, I'll just wait patiently this to become more mainstream and polished, like I did with Stable Diffusion and jump in when all the "beta testers" lost their time iron out the kinks. Business doing great without AI so far, so yea :)
@@StringerBell There's not much of a beta testing when it comes to agents. The only two limits are your ability to define your agents roles and the models capability.
And of course, you ability to set up your framework, which can be daunting.
You're probably waiting for a framework with a nice UI and straightforward installer. Which is going to happen very soon. But that will be a gpt3 moment just 1000x more pronounced
@@HakaiKaien Yep, I want agents to be able to actually operate PC's (not the stupid kind of Open Interpreter level)
Don't get it. How is this an agent? This just looks like prompting a local model :)
Thanks for a great educational video as always David. Personally for me i will be waiting for Elon's open source models to run locally. However what you showed can be incredibly useful for a lot of people. Thanks again and cheers 🤟🏻
What's the best way to do these open source models on a low budget. My laptop is 16 gb Intel i7 core. But anytime I open a model in LM studio it's hardcore struggling.
So i still can't do anything with my 8gb ram card ? Not rly free if you need spaceship pc to run llama
Hi, David. I would like to know if you can do a project to help promote a political candidate. I am trying to create something for a condidate a a spanish country.
You do recommend using multi agent frameworks? Ok, I am in the opposite boat. I think they are not the answer and ultimately less capable. So if you need some content and like a challenge how about this:
Create/use a multi agent framework to build a moderately complex piece of software, a service, and or application.
You can build upon a pre existing framework like autogen, crewai, chat dev, etc if you'd like and you can use any closed or open source LLM you want. Whether it be gpt4, claude, llama, anything. In terms of what is "moderately complex". Use common sense, a game of snake isn't going to cut it but maybe something like a chatbot connected to a vector store where the end user can provide documents to ingest then query the chatbot against those documents. That I dont think is too difficult but hard enough that it would prove multi agent ability.
Can you please make a video on how to make agents that will have capability to call multiple APIs like GET, POST requests and display the data from the response
Great video. Thank you. But what if I want to use two agents with two different models? Could you please show how to do this? Using specialized models for each agent is powerful. Thank you!!
nothing in this video is working. Could anyone succeed after following instructions step by step in this video?
No
It seems you have created a llama environment. Should this not have been show
glad u r on PC. iOS is meh
The MacBooks with the M2 and M3 chips are helpful. You won’t need a fancy NVIDIA graphics card to run these local LLMs at a good speed with Apple Silicon Macs.
God bless you, man! I am going to recreate your example project just to get a feel for using Llama-3 with Python, then I will branch out from there. I am saving money to sign up for your course!
Cool! Gave me an idea of creating a simple 2D virtual environment for like 4-5 agents and give them "work titles" and let them walk around by outputting directions, for example Move x+1, and let them communicate and do stuff that actually update the state of the environment and their individual current coordinate positions inside the environment.
thats a great idea. You could give them a proximity weighting so they can only communicate with others nearby. I wonder if they would huddle or spread apart like gas?
@@StephenHicks1 yeah would be cool to find out! I guess you have to crunch current ”map”, other agents positions, the current agent position, some ”personality” info, some instructions what to output if the agent should take action, like walking or talking and some history of past events, all into the prompt for each query. Seems simple in theory.
This is just a prompt chain. Not an agent …
Glad you’re keeping it local. 💪🏽
Hey guys! I'm immersed in the study of AI agents and I'm curious: would it be viable to build an agent that prospects customers for freelance professionals? I envision a system capable of exploring Instagram in an automated way, identifying potential customers and even starting conversations to schedule sales meetings. Is it possible to develop such AI agents? If so, do you know of any videos on UA-cam or any mentors that explain how to create AI agents to automatically prospect customers through Instagram? Is there something like this in development or is this an idea for the future of AI?"
Is there any way today to create and use ai agents while not being a regular or competent programmer who utilizes code? I'm familiar enough to understand code, generally, but no recent experience actually coding, it's been years. Seems like agents require a good bit of understanding of not only coding but other integration mandatory tools like python and whatever else... Is there any option to create useful ai agents today without programming?
+1 😉
can i build an ai agent that write personalized first line for my prospects doing the web research?
can you make this agent surf the web without api?
this is crazy.. glad i found ollama! not much python required to split it up and get agents going.. would be awesome to see tavern card implementation!!
For some reason I could not get the code to run calling for an "No module named 'ollama'" Even after installing ollama.
Love the glasses amazing insight
how do you write the response to a file !
ask chat gpt, he knows basic stuff like how to write a text variable into a text file
If the agent is giving wrong response how do we train it ? any help
I was trying to use phi-3-mini model on CrewAI and I was having problems because the model wouldn't understand the prompts from the previous agent to the next. When changing the temperature to lower, it sometimes worked, but it's not consistent.
Now I'll try writing from the scratch like shown in this video. Thanks for the tutorial!!
Create ai agent build for seo purpose
Make a video on how to earn money?
This isn't an agent, This is still a little bit of python script with an LLM lolz
What is an agent in your opinion?
Thats the point… lol
What are the characteristics of the pc that enable me to run llama3 locally ?
Almost any modern one would suffice. With 8 GB ram you should be able to run the quantised 8B model. To run it fast, you need a 6 GB GPU. A nvidia 4090 runs it at about 80 tokens per second, my laptop with a 3060 runs 33 tokens/second. Very useable. The 70B model requires 64 GB to run, but you will get only 1-2 tokens per second, even on a fast pc, because it does not fit your GPU memory. (you can also run 2-bit quantisation with 32 GB, but quality will be worse)
thanks
These are helpful, thank you! Just a small suggestion, it’s always a good practice to create a virtual environment for each new project! That way you won’t have to deal with conflicts, if ever..
Very awesome!
what is an agent?
it's just an llm ?
It's an llm with tools. Like they can scan the internet for new news stories and summarize it, or it can build a website, or it can allow you to chat with pdf files. LLM + tools
@@Educationsupport thanx, i have an other question, for example i build a agentic RAG application, this application has multiple LLMs working together (router, grader, generater, hallucination_checker, etc...) is every single LLM are called Agent or the whole application is an Agent ? (because i saw an information that agents break a task into multiple tasks).
also the chat prompt template for each LLM in the application, is it considered as prompt engineer ?