Outstanding project and great video, thank you so much for helping me learn in my living room 🙂🫡🙏 the feedback loop is way more powerful than many other frameworks. Genius
Good review! I think it would be beneficial if you walked through the source code in another video. Anyway ist funny that over the last two years, we have seen this move from over-bloated "AI agent" libraries to this 1000-line "code-interpreter"! So, it all comes back to code execution! This means that soon, when a model can natively execute code, we will not have any of these issues. Similar to how CoT or "think step by step" happens natively with reasoning models, eventually, code execution will become like that, and then we will no longer have any of this.
The first one is wrong, all ressources including including Wikipedia said a Leopard run at 58km. No need for agents, Claude and ChatGPT give the right answer directly but the DeepSeek V3 chat give a wrong Leopard speed at 90km. I tell it give the wrong answer and he agree and apologize for the mistake. We should find a way to tell the agent to use or not use complex agent. And in my rare test until now, i just see the limit and weaknesses of Deepseek. Most big LLM found the right answer directly, even Mistral and Llama 3. 8b!! .
Can you make a video on using small LMs to understand and generate code in local system. The use case here is that a lot of your audience may be developers, a majority of them may not have the liberty to upload their code base to frontier models and ask it to generate code for adding/explaning functionality.
yes, absolutely I d love to learn how to use it with ollama, then we can create our internal use cases agents.
Video coming soon.
caught me off guard when it changed its query and tried decomposing queries. the self heal rocks
These things can really surprise you some times :)
Outstanding project and great video, thank you so much for helping me learn in my living room 🙂🫡🙏 the feedback loop is way more powerful than many other frameworks. Genius
I agree, its a really neat implementation. We might start seeing more and more frameworks with code agents and feedback loops.
More on smol agents please
Good review! I think it would be beneficial if you walked through the source code in another video. Anyway ist funny that over the last two years, we have seen this move from over-bloated "AI agent" libraries to this 1000-line "code-interpreter"! So, it all comes back to code execution! This means that soon, when a model can natively execute code, we will not have any of these issues. Similar to how CoT or "think step by step" happens natively with reasoning models, eventually, code execution will become like that, and then we will no longer have any of this.
What they refer to as reasoning is actually latency 😅
The first one is wrong, all ressources including including Wikipedia said a Leopard run at 58km.
No need for agents, Claude and ChatGPT give the right answer directly but the DeepSeek V3 chat give a wrong Leopard speed at 90km. I tell it give the wrong answer and he agree and apologize for the mistake.
We should find a way to tell the agent to use or not use complex agent.
And in my rare test until now, i just see the limit and weaknesses of Deepseek.
Most big LLM found the right answer directly, even Mistral and Llama 3. 8b!! .
basically you just read their docs and python notebook :)
Stop whining you got the video for free 😂🤦♂️
Please make video about training a model
Can you make a video on using small LMs to understand and generate code in local system. The use case here is that a lot of your audience may be developers, a majority of them may not have the liberty to upload their code base to frontier models and ask it to generate code for adding/explaning functionality.
Also a video could be on choosing where to which small LM.
great
Sir please make a video to build a movie ticket booking agent
? 😂bro use the ai
Extremely limited at this point