Agree, diving head first into Agency-Swarm and can't move fast enough. Had two opportunities turn up this weekend. His concept goes beyond agents, but an independent "Agency" to contain and coordinate multiple agents.
People said Tesla P40 was too old. I got 2 anyway and got awful speed. Then things evolved and I now get speed as fast as I can read. Agents and long term memory are the next things that will make me free from my ChatGPT dependance.
FYI - When you see that error when your first run agency-swarm genesis about readline, it tells you if you are on windows to pip install pyreadline3, just exit install and re-run.
You don't have to solve them on your own - while I did do that in the video - you can always use ChatGPT + Perplexity + Agency Swarm documentation + joining their discord and asking for help. By combining these 4 resources I don't think there's any problem that can stop you.
The next big thing is the combo of agents working into an OS-WORLD github operating system, that will be a game changer, hope you can figure out this 2
Would be great if Agency Swarm supported other LLMs aside from OpenAI. I find claude-3-haiku to be much more capable than GPT-3.5 and I would prefer to use that one. Works great with CrewAI, so I'm stick with that for now.
I can't decide which strategy to start with, Versen community or David skool, I am new to the field and looking for best strategy to create ai agent for business requirement. Any advise will be much appreciated.
If joining both is not an option, sign up for David’s school, classes are great, community is awesome, and you’ll get the overview and foundation you need to make the most of Versen’s system.
I would personally recommend just looking at all the openly available resources [youtube videos or tutorials from vRSEN or David or any other youtubers] to make sure you will stick to the program [as in put in enough effort to learn the basics and after you know enough, you can then decide if all the information available for free is good enough for you to progress or if you need to join David's skool or other community where you need to pay in order to continue. Personally, I am doing this myself.
Great videos, thank you! I have a small question. maybe you can assist with that. I enjoy writing short stories, if i upload ten short stories that I wrote and then ask the agent to create a new story, based on my style of writing . can it do that ?
I'm able to do everything except getting that chatbox to convert the image. When I place the file in the Files folder and run it, it repeatedly says It can't access local directories and asks me to upload it in the chatbox. But it doesn't work either way. Nice video btw!
Bro, the video's great and all. I believe you can think of a better example than converting webp to png via chat bots.... This is an over engineering and completely useless. There are much better use cases for AI agents.
I’m excited to join your community, New Society, within the next week. I’m committed to making this investment because I see a significant untapped market in my field. As a professional in civil engineering and landscape architecture, I know these industries are in dire need of innovation. Too often, outsiders like programmers who don’t fully grasp the nuances of our work create ineffective solutions. As an insider with deep industry knowledge, I have numerous ideas and questions about how to drive meaningful progress. We cannot ignore the powerful ecosystem around us. The ongoing environmental neglect poses a serious threat to global stability. Our world faces an environmental emergency that demands immediate attention. It’s crucial that we address these issues now. Please, do your research and stay focused on the real challenges at hand.
How's that amazing? You can ask gpt4 to do that and it wouldn't take you nearly as long. You're getting caught up in all the fluff like creating .md and or other documents that take little to no ability. Make it do something worth looking at. It's cool that there is an appearance of communication between the agents. Its a catchy thing. If we draw inspiration from nature, as we have always done. Then multi agents cannot be the answer. The human brain is the target. It is a single, efficient, and capable "agent". It is the epitome of intelligence as we know it. A multi agent framework could always be made more efficient with less agents until we arrive at one. Currently Multi agent frameworks are overly complex, dont work, and lack elegance. They're sloppy and noisy and ultimately do not produce more effective code. If you disagree with that can you please provide me a real life example of a multi agent framework actually making something beyond a simple task?
The human brain is actually a triune brain. Whenever you are in a thought process you can witness your “agency “ taking place, so the multi ai agent system is inspired by nature.
1. It take a minimum amount of creativity to find a use-case that ChatGPT cannot do - all you need is a single API and you've already given your Agents more capability than GPT-4 has 2. Your analogy to nature has a major flaw - a single human brain absolutely CANNOT compete with 10,000 human brains. 3. Just because you are not able to develop an Agent team that works doesn't mean others can't. VRSEN, the creator of Agency Swarm, literally does this for a living - and his company has 15 employees, most of them AI Agent Developers
@@DavidOndrej Thats a fair point about one human vs many in some instances. But I still havent seen an example of anyone making anything actually useful with a multi agent framework. I think that the multi agent idea is nice and it sounds cool and catchy and gathers a lot of hype. Ultimately due to the unintentional obfuscation of embeddings during the passing of context between agents they tend to fail after a certain point. I am just saying, if you believe in them its fair to then assume you have actually seen them do something worth believing in. Someone starting a company with employees isnt really going to cut it here. You can just prove me wrong by making a video or a series of them where youre able to successfully build something moderately complex with a multi agent framework. If anything youll get some content out of it. its a win win.
I like your attitude towards errors, publishing your videos with errors and their solutions as they would be encountered, but this may only serve to be very offputting to your viewers. Also, as an educator you should never appear clueless regarding any aspect of your work/material. Your profession has a long and proud heritage and your viewers take you VERY seriously. Perhaps this video should be re-done.
It helps beginners see how to trouble shoot and fix errors within this system though it would be better if he compiled it wothout errors tested it then posted the video but as someone new to all this i must say it has value and has helped me better understand reading errors
🔥 LAST 24 HOURS! Get a custom AI strategy made just for you: www.skool.com/new-society
What form does it take? Is it an action plan ? A discussion with you ?
Agree, diving head first into Agency-Swarm and can't move fast enough. Had two opportunities turn up this weekend. His concept goes beyond agents, but an independent "Agency" to contain and coordinate multiple agents.
People said Tesla P40 was too old. I got 2 anyway and got awful speed. Then things evolved and I now get speed as fast as I can read. Agents and long term memory are the next things that will make me free from my ChatGPT dependance.
Same with my m40
Can you do a video on these? Loading llama 70b for example? They are pretty cheap so would be super interesting
@@lovol2 Llama 3 70b Q4K_M is really great at coding. Using 2 P40 I get 5.2 t/s with 8k context
What models are you loading and are you using gguf or other file format. Tokens per second?
@@MagagnaJayzxui I've been using DeepSeek Coder 33B Q8 and now LLama 3 70 Q4. Always GGUF as P40 only do int8.
21:54 that "oh-!" was hilarious 😂 great video David!
I appreciate you showing your work and how you overcame the errors, very helpful for people learning 🎉
69k subs, congrats! Looking forward to joining your community very soon!
FYI - When you see that error when your first run agency-swarm genesis about readline, it tells you if you are on windows to pip install pyreadline3, just exit install and re-run.
How do you like Agency Swarm compared to autogen and crewai? Does it have any standout features or are they all about the same?
oh my GOOOD the super advanced agency swarm converted a image to PNG, OH MY GOOOD THATS SO ADVANCED USE CASE 21:57
Man I like the idea that you are making a AI Agents for dummies course but how in the hell am I going to be able to solve these errors on my own?
You don't have to solve them on your own - while I did do that in the video - you can always use ChatGPT + Perplexity + Agency Swarm documentation + joining their discord and asking for help.
By combining these 4 resources I don't think there's any problem that can stop you.
thank you so much for your efort. i would love to watch a livestream of you bulidung agents.
It is great advertising for how Agents will create change.
The next big thing is the combo of agents working into an OS-WORLD github operating system, that will be a game changer, hope you can figure out this 2
Should we first create a virtual environment before installing the packages?
Would be great if Agency Swarm supported other LLMs aside from OpenAI. I find claude-3-haiku to be much more capable than GPT-3.5 and I would prefer to use that one. Works great with CrewAI, so I'm stick with that for now.
Yeah... that's perhaps the main downside of it. There's no support for open-source LLMs either
@@DavidOndrejI'll try to fix it for you guys.
Give me some time. It's shouldn't be that hard.
@@PseudoProphet do you have a github account?
I can't decide which strategy to start with, Versen community or David skool, I am new to the field and looking for best strategy to create ai agent for business requirement. Any advise will be much appreciated.
If joining both is not an option, sign up for David’s school, classes are great, community is awesome, and you’ll get the overview and foundation you need to make the most of Versen’s system.
I would personally recommend just looking at all the openly available resources [youtube videos or tutorials from vRSEN or David or any other youtubers] to make sure you will stick to the program [as in put in enough effort to learn the basics and after you know enough, you can then decide if all the information available for free is good enough for you to progress or if you need to join David's skool or other community where you need to pay in order to continue. Personally, I am doing this myself.
Looks interesting, but CrewAI let's me use Ollama. Much cheaper to run.
That is the main downside of Agency Swarm... it doesn't allow open-source LLMs
hey, my chatgpt doesnt have gpt4, and i wonder if there is an way to change it to gpt3.5 as i couldnt even get pass the api key stage
love the vids thanks for all the hard work
Great videos, thank you!
I have a small question. maybe you can assist with that.
I enjoy writing short stories, if i upload ten short stories that I wrote and then ask the agent to create a new story, based on my style of writing .
can it do that ?
yes
Спасибо!
Does this work with the pplx api too?
I'm able to do everything except getting that chatbox to convert the image. When I place the file in the Files folder and run it, it repeatedly says It can't access local directories and asks me to upload it in the chatbox. But it doesn't work either way. Nice video btw!
Requesting similar vids on GPT Researcher and GraphRAG
How much does the api used here cost??
0.96 USD
what's the difference between AI agents and AutoGPT?
"Dear Agency Swarm support.
One of my agents is being really dumb. It keeps saying I didnt upload the file. These agents are stupid."
😅😅
Bro, the video's great and all. I believe you can think of a better example than converting webp to png via chat bots.... This is an over engineering and completely useless. There are much better use cases for AI agents.
thaks for these free vdeos brother
At this point I think i've covered 90% of my goals with just my ai. Lol
Thanks for the video, 🎉
This is nice but I’ll just use the assistants api directly
100% ahead
I’m excited to join your community, New Society, within the next week. I’m committed to making this investment because I see a significant untapped market in my field. As a professional in civil engineering and landscape architecture, I know these industries are in dire need of innovation.
Too often, outsiders like programmers who don’t fully grasp the nuances of our work create ineffective solutions. As an insider with deep industry knowledge, I have numerous ideas and questions about how to drive meaningful progress.
We cannot ignore the powerful ecosystem around us. The ongoing environmental neglect poses a serious threat to global stability. Our world faces an environmental emergency that demands immediate attention. It’s crucial that we address these issues now. Please, do your research and stay focused on the real challenges at hand.
How's that amazing?
You can ask gpt4 to do that and it wouldn't take you nearly as long. You're getting caught up in all the fluff like creating .md and or other documents that take little to no ability. Make it do something worth looking at. It's cool that there is an appearance of communication between the agents. Its a catchy thing.
If we draw inspiration from nature, as we have always done. Then multi agents cannot be the answer. The human brain is the target. It is a single, efficient, and capable "agent". It is the epitome of intelligence as we know it. A multi agent framework could always be made more efficient with less agents until we arrive at one. Currently Multi agent frameworks are overly complex, dont work, and lack elegance. They're sloppy and noisy and ultimately do not produce more effective code. If you disagree with that can you please provide me a real life example of a multi agent framework actually making something beyond a simple task?
The human brain is actually a triune brain. Whenever you are in a thought process you can witness your “agency “ taking place, so the multi ai agent system is inspired by nature.
I'll just ask Llama3 in Groq directly and it writes functional code. It does Python, Javascript, React Native very well.
1. It take a minimum amount of creativity to find a use-case that ChatGPT cannot do - all you need is a single API and you've already given your Agents more capability than GPT-4 has
2. Your analogy to nature has a major flaw - a single human brain absolutely CANNOT compete with 10,000 human brains.
3. Just because you are not able to develop an Agent team that works doesn't mean others can't. VRSEN, the creator of Agency Swarm, literally does this for a living - and his company has 15 employees, most of them AI Agent Developers
Hive mind = agency.
So bees and ants for start 🙏
@@DavidOndrej Thats a fair point about one human vs many in some instances.
But I still havent seen an example of anyone making anything actually useful with a multi agent framework.
I think that the multi agent idea is nice and it sounds cool and catchy and gathers a lot of hype. Ultimately due to the unintentional obfuscation of embeddings during the passing of context between agents they tend to fail after a certain point.
I am just saying, if you believe in them its fair to then assume you have actually seen them do something worth believing in. Someone starting a company with employees isnt really going to cut it here.
You can just prove me wrong by making a video or a series of them where youre able to successfully build something moderately complex with a multi agent framework. If anything youll get some content out of it. its a win win.
I like your attitude towards errors, publishing your videos with errors and their solutions as they would be encountered, but this may only serve to be very offputting to your viewers. Also, as an educator you should never appear clueless regarding any aspect of your work/material. Your profession has a long and proud heritage and your viewers take you VERY seriously. Perhaps this video should be re-done.
It helps beginners see how to trouble shoot and fix errors within this system though it would be better if he compiled it wothout errors tested it then posted the video but as someone new to all this i must say it has value and has helped me better understand reading errors
💣
Seems like all these tools still suck and you can't actually do anything useful with them yet...
finish the video
@@khalifarmili1256 i mean, working through a bunch of strange errors just to get a webp converted to png -- I don't find that actually useful.
а ты че побитый? дрался?
1st
Hy David 💪thanks for your hard work!
Love your tutorials and i try to get along with it 💻👾
Unfortunately, I can't afford "New Society" 🙈😊
Same for me.