This is why I subbed. Jason is giving us the Agent content we need 💎 was just looking for this

Awesome! Thanks so much for creating this timely, clear, concise practical how to video!

Would love to see AutoGen + LangChain used: - to write long form content like books - to read code base and implement new feature. - Make agents to research and Viz data etc. I know there are many examples of these but not yet with AutoGen.

@atchutram9894

8 ай бұрын

I would like it to read large code base to assist human developers.

@charlieburton9610

7 ай бұрын

I'm sorry if this is a noob question but what role does langchain play? Won't a multi agent framework already have 'memory'? I even saw someone talk about 'recursive agents' -> agents the decide which agents are needed and spawn them. Again using Langchain. So what is its (Langchain) purpose

@J3R3MI6

7 ай бұрын

@@charlieburton9610 no worries I’m very new to coding and AI in general (just started during pandemic), but I think something like Langchain or GPT-Index might still be useful for processing very long text (until Open AI expands the context window) but you might be right about AutoGen already having that capacity.. I’d have look a bit more

Awesome content as always! Keep doing this great job Jason, thanks! Always a must see.

This is the best and most useful video about AutoGen. I love how you always have real world use cases.

Awesome video man. Quick installation tutorial, along with great example use-cases of you actually testing it out. Thanks! I'm getting started on my own projects now

Great info, the background music was stressing me though.

Thanks for posting this. You’re such a great resource in this space!

This is really impressive and practical. I love the soundtrack, too....gets intense as the complexity increases

Amazing, thanks Jason for bringing light to this.

Your tutorial is simple but informative. Thanks for the wonderful tutorial !

You're my favorite AI channel man. Keep them coming!

I don't ever comment on youtube but this absolutely deserves it. 11/10

Thank you for your amazing vids. I am learning a lot and will look into this. This looks really promising.

Awesome Vido, I can not wait to try that out. Implementing a speech to text and text to speech function and streaming on top of this could make it much more immersive, especially when creating a personal assistant agent

Hi Jason, could you think about teaching how to build a recursive agent? It seems like a logical next step to just create one agent that is capable of writing as many new agents as it needs to complete a task. Just have it read in it’s own initiation script along with the documentation for orchestrating group chats. I’m not sure how best to synchronize memory and implement version control. I think this should make these even more powerful.

@Wanderloots

7 ай бұрын

this is what i've been wondering about as well. Kind of like having the manager agent assemble its own team

What a great introductory video🤗

as always, great content! quick question, how do you combat rate limits being hit, imposed by specific openai model version without compromising quality of response by going into lower gen models?

Bro you're doing amazing work, keep it up!

Amazing intro to autogen! Tried metagpt, chatdev, autogen; Autogen really feels like the framework where i can have lots of control

I would love to see the lead gen in action. Thanks for your hard work !

Amazing video, can't wait to see what you come up with next!

@Lucky9_9

7 ай бұрын

Wow. Did not expect to see my favorite app comment on one of my favorite AI youtubers :D

always get these . after error message sayin i used 10000K per minute and ask me to try again later. Process finished with exit code 0 but watching your video make more sense now.. your are the Goat.

Excellent post as always.🎉

I hate that I have shiney object syndrome, but I just can't get enough of your videos and imagining what could be possible with this.

🎯 Key Takeaways for quick navigation: 00:00 🧩 Introduction to Multi-Agent Frameworks 00:54 🤖 Microsoft's AutoGen Framework Overview 02:05 🚀 Getting Started with AutoGen 03:44 🔄 Iterating and Refining with User Feedback 05:23 👥 Coordinating Multiple Agents with Group Chat Manager 08:55 🔍 Advanced Use Case: Content Generation Pipeline Made with HARPA AI

Great content, thank you for sharing!

very awesome , thanks for your content

Great video thank you Jason

Bro you are such a beauty! This is so juicy. Appreciate the tutorial!

I liked the video without even watching it! Now, I'm watching!

Thanks for share. It is great content!!

Thanks Jason!

Thank you ❤

Love your work, it would've been ideal if the autogen could handle the pipeline flow creation. The manual coding setup seems a bit counterintuitive when the aim is to leverage AI for automation. :)

This is crazy to imagine all the things we will be able to do with multiple agents!

The best value on KZread

@getzkillz66

8 ай бұрын

he is growing fast!

this is absolutely amazing. listen, I tried the content generator on a different topic - a tutorial about CSS - but it seems that when the content has HTML or CSS samples in it, the agent tries to run the HTML or CSS code and it gets an error which throws it in a loop. Is there a way around this? The HTML and CSS code is actually part of the blog post, so not something to be executed

What a great video! I would have loved to subscribe to your channel but then realized I'm already subscribed LOL!

Interesting. I'm already doing something similar in my article generation cloud function. So autogen is just an easier way to build such an iterative generation operation?

Excellent!

excellent video. please make more 🔥🔥🔥

Possible to show us to use a local llm like codellama or llama2 with autogen and write the code for putting a simple ui using gradio on top of autogen ?

Just a couple of years ago I made my first from scratch python program to plot graphs from my investments, now I did it in five minutes!

Hi Jason, you produced a series of helpful videos for introducing AutoGen for different purposes. I am wondering whether you can have a video for using Gradio or other libraries to produce UI interface for AutoGen, so we can share the work with those who do not have hard-coding skills. Thanks.

@AIJasonZ

7 ай бұрын

good idea! which autogen use case you built? there are lots of frontend implementation options

I wish you also talk about the costs involved with using these and the costs for your own developments for each video in terms of LLMs & APIs usage.

i'm not smart so i always had a hard time how to use langchain etc so glad i found this channel

@Wanderer2035

8 ай бұрын

Yea me neither. I don’t even bother just let the AI figure it out lol

it will be nice if we see how the OAI_CONFIG_LIST Looks like for the content agent. you use 2 APIs? or only one? thank you again for you great work. it's really impressive.

Is the generated blog post up somewhere where we can check it out?

Hi Awesome stuff! How much it cost you to create the blog post? a few cents or more? and was it around 1500 words or less?

You are the best!

I followed along with the tutorial thank you for teaching me some new skills. I then had them draw me a photo of a cat. I gave them feedback and now I have python code that draws cats.

@Lucky9_9

7 ай бұрын

This is truly what technology is made for. Excellent work :D

Is that pycharm youre working in to download the autogen?

Can we use any fine tuned llama2 model in this setup? If so, it will very helpful if you make tutorial on that. Thanks for making such high quality video tutorial.

amazing

I’d like to see info on adding more personality/backstory/specific knowledge persona style info to the agents to create a stable staff team with different skills, ‘qualifications’ and perspectives. Eg can they have a file about themselves which they reference? Also a comparison of Chatdev and Autogen for this type of approach would be useful

@adrianlemus2768

8 ай бұрын

Do you envision this would allow them to put a specific spin on their work? Or what reason do you have for thinking this would be a benefit?

@_SimpleSam

8 ай бұрын

@@adrianlemus2768 It definitely IS useful. Every token of backstory/character/history will modulate the response. This can even include things like emotional content. One of my go to solutions to solve issues is to add small increments of detail to the prompt. ex. perfectionist, wise, well-read, genius, fully focused, etc.

@mickelodiansurname9578

7 ай бұрын

here is something I found about prompting. So there are the usual tricks like "In a step by step fashion" or "take a deep breath and carefully...." but I found it you start in a system message giving the AI an persona such as : 'You are Angus "Mac" MacGyver and are fascinated with coding, your python skills match your persona, put it together and remember that your life could depend on it working... lets get cracking right now. MacGyver, your task is writing python code to set up a call to the google api to get information on how to use 'moviepy' . First you'll probably need to think up a few searches for google, Here are some handy suggestions you found while looking around. 1. "moviepy library Python tutorial" 2. "Getting started with moviepy Python" 3. "moviepy documentation" Good luck MacGyver we are depending on you.' So yes it sounds funny right... using a 1980's TV hero as the persona... but think about its training data and what it pattern matches to MacGyver and try it, watch how it makes no mistakes in the code, how it writes a loop nice and easy without prompting... so it is basically a MacGyver Coder agent.

can you please also share the repository for the above code. And can i use Ollama instead of gpt?

Nice content.

This is great. Can this help in performing complex queries on tables in large postgrs db using natural language without sql?

now to use this with an open source model like mistral, or some other model with a much larger context window.

Another great video, thanks Jason! If you think about it, multi agent conversations is looping though participants over and over. Its super simple. The codebase of Microsoft's Autogen is a dumpster fire unfortunately (no hooking into simple under the hood events, no simple polymorphism to extend the functionality, 150+ lines long functions, etc.) I wrote my own simple conversation logic, at it works just as well.

@digitalsoultech

8 ай бұрын

Care to share?

why those projects always come with openai? is there a way use this offline with llama etc?

I have an error on RATE LIMIT with the GPT api. Anyways to solve this?

Opportunity - Agents as a Service. Wrap Agents curated for a specific niche like lawfirms in a SaaS like platform and charge monthly. What do you think?

@brisonvsn

8 ай бұрын

Opportunity - Agent Agency. Help companies create Agent offices

Can we implement autogen using open source models without open AI? If yes how can we do it?

i would love to use this with a better UI to test this out

Damnn your like GOD !!! Love your stuff n realy want to dive deep in to it. But i dont know any thing about coding. So its gonna be a ruff one for me. But im gonna watch some more of your videos n than try to install it. Would realy like to make a ai company. I work with kids so i would love to show them this.

@dailywisdomquotes518

8 ай бұрын

Hey, Loved your enthu, i'm also working on AI start up, would love to connect

👍

How much was the token used? I tried for my creative work and it would use the token around 6-7 dollars per work which is crazy (gpt-4)

What program is Jason doing all of this on?

@rabanal_josh64

7 ай бұрын

Vscode

Great one

finally! :)

Your content is incredible! Thank you for such a clear and thorough explanation. If I didn’t know any better I might think you are a GPT5 agent 😂. For autogen, would it be possible to link this framework to a private database with embedding so that we have a combination of internal and external research in the report?

@mlopolis

8 ай бұрын

I think you could do it. Just add another agent to query the local database and get the information you need using cosine similarity. The problem is, you'd still send the local info to OpenAI, so it won't stay private.

remember to protect your API key

Hi jason, could you do a video on autogen using llms

The future is now 🤯

i really want to see read code base and implement new feature but with telemetry based prompt and automated deploy, like automated frkn SAAS business.

Awesome content but the background music was really stressful 😢

have people tried this with other companies that are starting from scratch and NOT trying to build a video game?

🎯 Key Takeaways for quick navigation: 00:00 🤖 Introduction to multitasking frameworks - Challenges in providing feedback to agents and limitations in the number of agents in a conversation - Introduction of Autogen as a multi-agent framework that addresses these challenges with user proxy agents and group chat manager. 01:22 🛡️ User proxy agent and human feedback points - User proxy agent acts as an intermediary between users and other assistant agents. - Allows users to provide feedback during the process, enabling greater control and iteration. 03:01 💼 Group chat manager for collaboration - Group chat manager facilitates coordination between multiple agents in a conversation. - Enables the introduction of additional agents to provide different perspectives and expertise. 03:44 💻 Example: Simulating a conversation between an assistant and user proxy agent - Demonstration of a basic use case where user proxy agent interacts with the assistant to generate stock price charts. - User feedback and iteration to enhance the generated output. 04:26 📋 Demonstrating a coding agent with group chat - Creation of a coding agent involving a user proxy agent, coder, and product manager. - Coordinated conversation and iteration for building a Python game. 05:51 📊 Use case: Content generation pipeline with multiple group chats - Illustration of a content generation pipeline using Autogen's group chat functionality. - Separate group chats for research and content writing tasks. 06:47 🔍 Creating a research function with function calling in Autogen - Demonstrating the creation of a research function using the SerpAPI and BeautifulSoup libraries. - Exploring the use of function calling within Autogen to chain different functions together, such as search and scraping. 08:13 🤖 Defining an agent with function calling and setting up human feedback points - Defining a legendary model config for a researcher agent with different functions, inputs, and descriptions. - Creating a user proxy agent to provide human feedback points based on specific keywords, such as "terminate." 09:37 📊 Triggering a conversation and retrieving research reports - Initiating a conversation between the user proxy agent and researcher agent to retrieve research reports. - Demonstrating the retrieval of the final research report using a specific message sent to the researcher. 10:20 💼 Creating a content generation function with group chat - Creating a content generation function involving multiple agents, including an editor, writer, and reviewer. - Setting up a group chat for these agents to collaborate and iterate on the content creation process. 11:46 📝 Triggering a conversation for content generation - Triggering a conversation between the user proxy agent and content assistant to generate blog content. - Showcasing the workflow involving the research function, content writing, and feedback iteration using various agents. 12:27 💡 Exploring other use cases and possibilities with Autogen - Discussing the flexibility of Autogen in creating different use cases like lead generation and marketing campaigns. - Encouraging further exploration of Autogen's capabilities and its potential applications in various scenarios. Made with HARPA AI

You left your serper api key in the repo

Hey I released a Langchain tool bridge so we can use the toolkits natively, its in the AutoGen repo as a PR, thought you might find it useful

@AIJasonZ

8 ай бұрын

Nice, do you have link?

No one’s going to point out how he kept repeating that we had to call TERMINATE to ask for human input. I see you trying to fool us SKYNET.

I don't know why, but for some reason...It's not saving any code anywhere...

I would like to use this with my Crypto bot to do some research and the agent turned code of the bot and buy the tokens, i need the agent to change the code of my bot every time when i have opportunity, thank alot

WE ALL NEED TO SEND HIM THANK YOU CARD

Realizing that LLM's aren't the holy grail and developing frameworks to "let people back in the loop" is ironic.

What is the benefit of breaking a task down into agents? Wouldn't AI's in the next year or so just be able to perform everything itself?

@digitalsoultech

8 ай бұрын

No, LLMs think better by breaking it down into piece by piece. This is the way of how the future of AI is and will be.

@adrianlemus2768

8 ай бұрын

I think in the interim they have a difficult time with nebulous requests. Maybe in a year or more it’ll get to the point where it’s redundant.

Please explain make a longer video explaining it further I tried following you and everything was working until I got stuck at 3:36. Instead of boom the results I got boom a long list of errors even though I followed everything and the autogen install guide and it seems like I can't run anything like the graph showing up or the pingpong game showing up.

@helix8847

8 ай бұрын

Ask ChatGPT to help you solve the errors?

@bobbob-mi6pq

8 ай бұрын

@@helix8847 I've spent nearly a week trying to resolve issues, but I still can't get it past the first message or just an error in the first message

@KamilKaczmarekSolutions

7 ай бұрын

@@helix8847 always works :))

Is t also post to put a agent in that is a bit negative to ideas an realy needs to be convinced by the other agents. Just so the idea would be extra solid? Greetings n stay godlike

in the near future a company's product will be an agent fine tuned on their proprietary knowledge that you can use in an autogen like group chat.

How long before people start moving away from Autogen? Only a matter of time.

please share the codes shown in video 🙏🏻🙏🏻🙏🏻

The amount of alpha this channel shares is off the charts.

music is a tad too dramatic, otherwise great video

Amazing video and not so amazing music 😂

@AIJasonZ

8 ай бұрын

Hah feedback received; was wondering whether I should keep the music actually

@EnesKaraboga

8 ай бұрын

@@AIJasonZ I would vote to totally get rid of it bro

@Mangini037

8 ай бұрын

Music needs to go Jason, your content is too good. Would love to see more AutoGen content mate. This is the best tutorial on AutoGen so far and explaining real world cases.

@AIJasonZ

8 ай бұрын

Got cha, thanks a lot for the feedback team, will update it for upcoming videos!

@osten222312

7 ай бұрын

TERMINATE

This is the biggest employment killer since ChatGPT. Whole companies will soon be pure AI

@brisonvsn

8 ай бұрын

@digitalsouthtech Opportunity - Agent Agency. Help companies create Agent offices .

@digitalsoultech

8 ай бұрын

@@brisonvsn right so the only jobs left soon will be the ones that transition companies workers into automation agents

@brisonvsn

8 ай бұрын

Not the only ones but that's a future business opportunity that could blow up.@@digitalsoultech

Looking forward to become unemployed.

I had to double-check to see if I had the playback speed accidentally on 1.5x.

your videos are very educational, but can you slow down the speech, it is extremly difficult for us non english speakers. sorry if it sounds rude and impolite.