With the automatic audio dubbing from TH-cam /Google you hear a synthetic voice in your regional language. To hear my original voice in English, switch to "Default" or "English" in the settings. Thank you.
This looks extraordinary! It feels like a whole new level of abstraction and sophistication for making agents. I think it’s going to be revolutionary, thanks for the clear presentation!
Amazing! Beautiful! Im already neck deep in it and I love it. Never going back to the old ways, just moving forward. Thx again! SmolAgents and SLMs for the win!
A month ago I was humbling about this same vision, your _Lectio Magistralis_ showed me the practical implementation of the dream I had. Thank you and thanks to HF and its crew who published this innovative thinking about how AI can grow whith LLMs *and* the already mature programming libraries ecosystem. Without wasting time, energy and money into even more fat LLMs with all the human knowledge into them, but so compressed that it's inevitably lossy (imprecise, hallucinated). Happy new year from Italy and thank you again from your lessons.
Please let us know if you find the swarm model dynamic conversion for smolagents. Before this, I thought pydantic ai might be the improved answer, but with some testing of the code agent power of smolagents, i think that may be "the one". Both swarm and py-ai talk about "agents" really as being functions or "micro agents", so would love this built in handoff capability integrated into smolagents for a swarm with a coding agent driving micro-agents, or driving a knowledge graph process through code. PS... first coding agent I used was using youtube libraries, not ML ones. Sacrilegious i'm sure. :).
Yeah this is cool. And yeah, Python, because 1) tons of libraries, and 2) LLMs are good at Python coding (well, at least better than many other languages). It will be quite wild when an agentic system can dynamically call the appropriate library for symbolic processing.
The manager agent reminds me of the orchestrator agent from other frameworks. We could also imagine having agents in a team to optimize agent performance, spawn or remove agents, or to critique the outputs from the team of agents. That should nudge the team in the direction of MoE or CoT approaches. What I’m also wondering about is whether there is a scenario in which teams are defined with properties rather than individual agents. This would mimic biology in which hierarchies of teams form components of higher level teams. That’s an alternative to using a manager agent as an intermediary to high levels.
Your video is arriving in Brazil, but if it is not dubbed into Portuguese there is no way to follow it, as the listener gets lost in the process of collecting information.
@@haroldpierre1726 yes, Google thing. I'm French but I prefer in English because some words are not well translated yet. For sure, they will improve the quality sonner or later. This is a good start BTW.
With the automatic audio dubbing from TH-cam /Google you hear a synthetic voice in your regional language.
To hear my original voice in English, switch to "Default" or "English" in the settings. Thank you.
Please add support for arabic dubbing .Thank You♥
This looks extraordinary! It feels like a whole new level of abstraction and sophistication for making agents. I think it’s going to be revolutionary, thanks for the clear presentation!
Agreed, been waiting on a framework that gets it right and this is the first one that is aligned with how I’ve been thinking
This is going to make learning all this stuff so much easier, thank you!
Amazing! Beautiful! Im already neck deep in it and I love it. Never going back to the old ways, just moving forward. Thx again! SmolAgents and SLMs for the win!
A month ago I was humbling about this same vision, your _Lectio Magistralis_ showed me the practical implementation of the dream I had. Thank you and thanks to HF and its crew who published this innovative thinking about how AI can grow whith LLMs *and* the already mature programming libraries ecosystem. Without wasting time, energy and money into even more fat LLMs with all the human knowledge into them, but so compressed that it's inevitably lossy (imprecise, hallucinated). Happy new year from Italy and thank you again from your lessons.
Wow! Everything I’ve been working on and wishing for together in one video. Thank-you!
Please let us know if you find the swarm model dynamic conversion for smolagents. Before this, I thought pydantic ai might be the improved answer, but with some testing of the code agent power of smolagents, i think that may be "the one". Both swarm and py-ai talk about "agents" really as being functions or "micro agents", so would love this built in handoff capability integrated into smolagents for a swarm with a coding agent driving micro-agents, or driving a knowledge graph process through code. PS... first coding agent I used was using youtube libraries, not ML ones. Sacrilegious i'm sure. :).
Yeah this is cool. And yeah, Python, because 1) tons of libraries, and 2) LLMs are good at Python coding (well, at least better than many other languages).
It will be quite wild when an agentic system can dynamically call the appropriate library for symbolic processing.
The manager agent reminds me of the orchestrator agent from other frameworks. We could also imagine having agents in a team to optimize agent performance, spawn or remove agents, or to critique the outputs from the team of agents. That should nudge the team in the direction of MoE or CoT approaches. What I’m also wondering about is whether there is a scenario in which teams are defined with properties rather than individual agents. This would mimic biology in which hierarchies of teams form components of higher level teams. That’s an alternative to using a manager agent as an intermediary to high levels.
Fully agree with the combination of agents and graph systems as the way forward. Memory optimization in the teams of agents is largely unexplored.
❤
Your video is arriving in Brazil, but if it is not dubbed into Portuguese there is no way to follow it, as the listener gets lost in the process of collecting information.
I have no way of influencing Google's resources for dubbing my videos. I can't tell you when Google will activate it, or why not or when.
Is this audio track a TTS?
automatic translation by google, you may want to choose the original track in the video settings :)
@@Jvo_Rien Google is doing the voice? I listened to this in Italian and French. If this is a Google voice, it's really good!
@@haroldpierre1726 yes, Google thing. I'm French but I prefer in English because some words are not well translated yet. For sure, they will improve the quality sonner or later. This is a good start BTW.
Face reveal when
after lil bro gets oiled up
/s (sarcasm ofc)
Pon doblaje al español. Spanish audio please
They misspelled SMALL, those dummies!!
Add Hindi Audio Track like mr beast