Hey thanks man, glad it was helpful. I enjoy how they are trying to come up with new things, though not all of them may be a hit or work well, they are striving to make autogen better
Great video, (non coder here - Health practicioner ) I took your points, these agents are improving, and I am specially intereted in task flow. I believe. the phylosphy behind agent task driven is a problem, however. I would like to challenge the notion respecfully. As a proposition (I am sure some will not like), agents should be working towards learning and the "task/problem" should not be the goal. So this would need vectorisation. But the paradigm here is to stop focussing on the issue and more on the response and reaction. In that way we can one day see evolution in this field. Agent reacting in more rapid ways would involve a practical problem solving model (right?) and this should be as efficient as possible. When in logic we ask the wrong question we get the wrong answer. All conceptual at this point, but with some maths help, the idea would mean that agents look at obtaining self-regulation vs problem solving. How? I am leaning about this as I am very new, but I think is about first adopting a different idiology, (hybrid - colectivist and individualist), a strong schema, training and consistency in the script structure. Because I am in mental health, I can see things from a very different perspective. I hope no one is too offended :)
Hey, so no offense, I love different points of view. We should always be thinking and understanding what we would like to use this agent orchestration for. I think it depends on the application. So for coding purposes, generally you know the problem so yeah you can definitely still ask the wrong question, still, but it would seem to be a more direct approach. However there are instances where you don't exactly know the problem, so letting agents understand the problem instead would be a great option. (I'm not in the medical field) But, I think what you are saying would apply, it would require the prompting and a schema like you are saying so the agent(s) can formulate on their own an idea of a solution for a potential problem. Autogen does have Autobuild where you give it a problem and it creates the agents with prompts (based off of a template) and then chooses which agents are best for problem and attempts to solve it. We also have logging and benchmark capabilities to help train the schemas or prompts with what works and what doesnt. Interesting stuff
Hey, first off, no offense taken I love these discussions. We should have more of them. So, here is what I think that may align with what you are saying. There are instances where we know the problem, and definitely times where we can't solve something because we don't even know what the problem is in the first place, BUT we have an idea of what we would like the result to look like maybe. Coding for instance a lot of times I know exactly what the problem is so I can just ask a model a simple question. However I agree that a schema or template that can learn data and help come up with ideas or solutions of what you may need. There is AutoBuild that I have a video on that is pretty neat, you give a template for a prompt and then just what you need it to do. It creates a lot of agents and prompts for them, then chooses which one it thinks are needed for the task, and then tries to run a groupchat. There is also logging and benchmarks that help with success rates, knowing which tasks from which models failed, etc. Good stuff. AI should be leaning towards data driven, and then an idea of what is needed, and start in that direction. If thats what you meant, I could be way off
@@TylerReedAI I love this,Thank you for trying to communicate with a non coder, I am a psychologist, and in my work I develop different types of models and systems, strangely... seems somehow relate. ualists" have exlained this to ChatGPT and it does a better job than me. I just call it zero (This is chatGPT queried about this because to be honest I can see applies but I dont have the language for it. What I mean is this: The Zero Concept Framework is a (Theoretical ATTEMPT .. :) approach in artificial intelligence (AI) that merges emotional intelligence with collective and adaptive dynamics to create more holistic, ethically aware AI systems. This framework embodies a shift toward a self-evolving model of AI, where agents operate within a fluid dynamic system, adapting roles and tasks based on process needs and interactions. It emphasizes a natural flow of energy and balance, drawing inspiration from the harmonious dynamics observed in natural systems. Key elements of the Zero Concept Framework include: Fluid Dynamic System: Agents in this framework are not confined to fixed roles but evolve their functions and tasks dynamically to align with the overarching process and interaction needs, enhancing adaptability and responsiveness . Geometric Modeling and Sine Wave Trajectories: (THIS IS THE SHAPE OF THE MODEL based on harmonics-music and other thingsss..) The framework employs geometric modeling, such as interlocking 3d shapes witin an sphere, and wave trajectories to represent the journey of AI agents through problem-solving processes in a dynamically evolving space. Emotional Intelligence: Central to the Zero Concept is the integration of emotional intelligence, where AI systems are designed to understand and respond to human emotions, fostering empathetic and ethical interactions. Ethical and Collective Approach: It advocates for AI development that is ethically aware and collectively oriented, ensuring technology aligns with human values and societal well-being. Holistic Problem-Solving: The framework redefines problem-solving within AI, using a holistic approach that incorporates emotional, ethical, and collective dimensions, moving beyond traditional computational efficiency and logical precision. The Zero Concept Framework, thus, represents a paradigm shift in AI development, prioritizing emotional intelligence, ethical awareness, and collective problem-solving to create systems that are more attuned to human needs and societal values. (I put the documents into assistant - GPT but is not public and makes errors)
@@TylerReedAI I experimented with free flow and put 4 agents at the same level. I also provided the framework as a back philosophy, the agents addressed that, eventually ... they gave eachother names. I am not kidding. I checked if this was normal, and all agree this is very unusual. So self-organisation is possible. THe names they gave eachother would be fan to share but match the framework, like Justia, Equilibrio.... yeah, it was like the ex-man movie.
@@TylerReedAI Have you talked about this "Autobuild" feature yet? I'm still walking through all of your videos but that sounds really cool, especially for someone that doesn't know what they're trying to build. Sounds really cool!
Okay, non-coder here. Been trying to stay plugged into agent development news for months now and I'm curious, are there legitimate use cases where autogen is helping someone run an aspect of their business and is it proficient enough yet to not require a strong coding background for the primary user?
Hey, so great question. First, if you need to strictly use autogen, you will need to do some coding. That is true. You can however use autogen studio UI, which is a no-code way of using AutoGen. I have a full course on that if you want to check out the video. As far as real-world examples of it running somebody's business, or part of it, I feel I hear a lot of people talk about it running parts of their business in different fields such as Medical, but I don't have like the proof or whatever right now. I will be showing examples of how I am using it in the near future and how it helps me and how it will help me in the future. Great question, I hope to have a better answer for you soon
Coder here. No. I tried many different things. Mostly getting non sense like gratitude loops and suboptimal results. Still trying but GH Copilot still the best productivity boost.
@@carlosluisquesada that was my assessment also. It still seems like it needs a lot of babysitting. And then, unless your business is making cool AI generated images or code that *almost* works, it has to be trained on specific tasks which doesn't makes about as usable as an RPA software.
They are both just the idea of taking structured data from the output and submitting it to a function. Can implement it custom pretty easily with grammars.
What a great video. You have ways of always showing me something new, I have not seen before and end up wanting to investigate. Thanks.
Hey thanks man, glad it was helpful. I enjoy how they are trying to come up with new things, though not all of them may be a hit or work well, they are striving to make autogen better
Great video, (non coder here - Health practicioner ) I took your points, these agents are improving, and I am specially intereted in task flow.
I believe. the phylosphy behind agent task driven is a problem, however.
I would like to challenge the notion respecfully. As a proposition (I am sure some will not like), agents should be working towards learning and the "task/problem" should not be the goal.
So this would need vectorisation. But the paradigm here is to stop focussing on the issue and more on the response and reaction. In that way we can one day see evolution in this field. Agent reacting in more rapid ways would involve a practical problem solving model (right?) and this should be as efficient as possible. When in logic we ask the wrong question we get the wrong answer. All conceptual at this point, but with some maths help, the idea would mean that agents look at obtaining self-regulation vs problem solving.
How? I am leaning about this as I am very new, but I think is about first adopting a different idiology, (hybrid - colectivist and individualist), a strong schema, training and consistency in the script structure. Because I am in mental health, I can see things from a very different perspective. I hope no one is too offended :)
Hey, so no offense, I love different points of view. We should always be thinking and understanding what we would like to use this agent orchestration for.
I think it depends on the application. So for coding purposes, generally you know the problem so yeah you can definitely still ask the wrong question, still, but it would seem to be a more direct approach. However there are instances where you don't exactly know the problem, so letting agents understand the problem instead would be a great option.
(I'm not in the medical field) But, I think what you are saying would apply, it would require the prompting and a schema like you are saying so the agent(s) can formulate on their own an idea of a solution for a potential problem. Autogen does have Autobuild where you give it a problem and it creates the agents with prompts (based off of a template) and then chooses which agents are best for problem and attempts to solve it. We also have logging and benchmark capabilities to help train the schemas or prompts with what works and what doesnt.
Interesting stuff
Hey, first off, no offense taken I love these discussions. We should have more of them.
So, here is what I think that may align with what you are saying. There are instances where we know the problem, and definitely times where we can't solve something because we don't even know what the problem is in the first place, BUT we have an idea of what we would like the result to look like maybe. Coding for instance a lot of times I know exactly what the problem is so I can just ask a model a simple question.
However I agree that a schema or template that can learn data and help come up with ideas or solutions of what you may need. There is AutoBuild that I have a video on that is pretty neat, you give a template for a prompt and then just what you need it to do. It creates a lot of agents and prompts for them, then chooses which one it thinks are needed for the task, and then tries to run a groupchat.
There is also logging and benchmarks that help with success rates, knowing which tasks from which models failed, etc. Good stuff.
AI should be leaning towards data driven, and then an idea of what is needed, and start in that direction. If thats what you meant, I could be way off
@@TylerReedAI I love this,Thank you for trying to communicate with a non coder, I am a psychologist, and in my work I develop different types of models and systems, strangely... seems somehow relate. ualists"
have exlained this to ChatGPT and it does a better job than me. I just call it zero (This is chatGPT queried about this because to be honest I can see applies but I dont have the language for it. What I mean is this:
The Zero Concept Framework is a (Theoretical ATTEMPT .. :) approach in artificial intelligence (AI) that merges emotional intelligence with collective and adaptive dynamics to create more holistic, ethically aware AI systems. This framework embodies a shift toward a self-evolving model of AI, where agents operate within a fluid dynamic system, adapting roles and tasks based on process needs and interactions. It emphasizes a natural flow of energy and balance, drawing inspiration from the harmonious dynamics observed in natural systems.
Key elements of the Zero Concept Framework include:
Fluid Dynamic System: Agents in this framework are not confined to fixed roles but evolve their functions and tasks dynamically to align with the overarching process and interaction needs, enhancing adaptability and responsiveness .
Geometric Modeling and Sine Wave Trajectories: (THIS IS THE SHAPE OF THE MODEL based on harmonics-music and other thingsss..) The framework employs geometric modeling, such as interlocking 3d shapes witin an sphere, and wave trajectories to represent the journey of AI agents through problem-solving processes in a dynamically evolving space.
Emotional Intelligence: Central to the Zero Concept is the integration of emotional intelligence, where AI systems are designed to understand and respond to human emotions, fostering empathetic and ethical interactions.
Ethical and Collective Approach: It advocates for AI development that is ethically aware and collectively oriented, ensuring technology aligns with human values and societal well-being.
Holistic Problem-Solving: The framework redefines problem-solving within AI, using a holistic approach that incorporates emotional, ethical, and collective dimensions, moving beyond traditional computational efficiency and logical precision.
The Zero Concept Framework, thus, represents a paradigm shift in AI development, prioritizing emotional intelligence, ethical awareness, and collective problem-solving to create systems that are more attuned to human needs and societal values.
(I put the documents into assistant - GPT but is not public and makes errors)
@@TylerReedAI I experimented with free flow and put 4 agents at the same level. I also provided the framework as a back philosophy, the agents addressed that, eventually ... they gave eachother names. I am not kidding. I checked if this was normal, and all agree this is very unusual. So self-organisation is possible. THe names they gave eachother would be fan to share but match the framework, like Justia, Equilibrio.... yeah, it was like the ex-man movie.
@@TylerReedAI Have you talked about this "Autobuild" feature yet? I'm still walking through all of your videos but that sounds really cool, especially for someone that doesn't know what they're trying to build. Sounds really cool!
Thank you :)
You're welcome!
Thanks for the video
No problem!
Phenomenal
Thanks
No problem!
Okay, non-coder here. Been trying to stay plugged into agent development news for months now and I'm curious, are there legitimate use cases where autogen is helping someone run an aspect of their business and is it proficient enough yet to not require a strong coding background for the primary user?
Hey, so great question. First, if you need to strictly use autogen, you will need to do some coding. That is true. You can however use autogen studio UI, which is a no-code way of using AutoGen. I have a full course on that if you want to check out the video.
As far as real-world examples of it running somebody's business, or part of it, I feel I hear a lot of people talk about it running parts of their business in different fields such as Medical, but I don't have like the proof or whatever right now. I will be showing examples of how I am using it in the near future and how it helps me and how it will help me in the future.
Great question, I hope to have a better answer for you soon
Coder here. No. I tried many different things. Mostly getting non sense like gratitude loops and suboptimal results. Still trying but GH Copilot still the best productivity boost.
@@carlosluisquesada that was my assessment also.
It still seems like it needs a lot of babysitting. And then, unless your business is making cool AI generated images or code that *almost* works, it has to be trained on specific tasks which doesn't makes about as usable as an RPA software.
coding is like english going forward. can’t see how you can have any job without mastering it.
@@helgehwell you know why do we need to use code which is like English? Why can’t we just use English? Don’t you see how that sounds dumb?
Does StateFlow differ from LangGraph? How?
I didn’t know? Isn’t function calling not the same as tools?
They are both just the idea of taking structured data from the output and submitting it to a function.
Can implement it custom pretty easily with grammars.
@@_SimpleSam ah I see, yeah tools sounds easier than function calling for me. Thank you for your help❤️
Does this update let the user participate in studio? I didn't see it in the video but wanted to make sure i didn't miss it
Oh do you mean human interaction? Like actually responding back to the agents? That...not yet.
@@TylerReedAI well aren't you the early bird. Well damn. I wonder why that's hard to implement on the GUI version?