Thanks for this short tutorial. I'm still new into all these, but I love the fact that you were able to put all these and explained thoroughly within this short stipulated time.
The browser can also be considered an OS, for web based experiences - that runs inside another OS. In any case, Karpathy's idea is more of a concept of what the future may look like.
it's interesting, but i don't quite understand the purpose. Is it supposed to make apps easier to write? Hardware easier to add? As unpredictable as LLMs are , it sounds like chaos but maybe I just don't get it.
Can you imagine building a basic LLM-based OS kernel with just a terminal, where you would use prompts to build every single one of the applications? That'd be awesome! 😎👍
I don't get the purpose of this either. Everything you have shown can be done with Open WebUI and Ollama too thanks to the recent addition of tools and functions combined with LLMs finetuned for function calling such as the latest Llama3. Storing stuff in a vector database isn't new either, this just gets you an expert system since the actual difficulty is for an LLM based AI system to decide when to pull what in the context and how much while quickly getting rid of trash. Also on conceptual level an LLM shouldn't replace the kernel since like the book indicated this is just an abstraction layer to make the computer actually usable. The LLM should be more or less on the same level as the user. It may need an own memory management system though that could even be driven by the same LLM just like MemGPT demonstrated. So what actually would be helpful would be an LLM that has a good understanding about the operating system and a very direct access to it without trying to replace it. With this it should be able actually do administrative tasks such as detecting errors, checking logs, performing updates and installations even if there are difficulties to overcome. Finally it should be able to get actual work done. I'm wasting so much time with system administration while at the same current LLMs have such a deep understanding that should be able to do most of the tasks on their own.
be sure to check out Frdel/agent-zero it's like a less polished, more recursive version using agents that call agents and respond to their creator, even able to troubleshoot itself - to a point
No offense but this is presentation is not even close to be an OS. It's just like RAG or adding some tools in your app, so that llm can invoke a particular tool based on user's input, like when to hit a search api. If this app do task like "I have deleted my financial report of july, can you recover it from recycle bin?" or "can you install xyz game, the file is in the download folder?" then we could have something different here. Anyways your content is awesome and very helpful. Cheers
Definitely no offense taken. I can completely understand that. I also had the initial view like that. That's why for many months I didn't make the video about this. But then I thought it's a good start. You never know so made the video
@@1littlecoder cool and thanks for reply. Would've thought he mentioned much earlier as have seen others moot similar in regard of LMMs going back over a year or 2
Thanks for this short tutorial. I'm still new into all these, but I love the fact that you were able to put all these and explained thoroughly within this short stipulated time.
How is this an OS? How is it talking to hardware? Does it write its own drivers? This seems like just an app...
The browser can also be considered an OS, for web based experiences - that runs inside another OS.
In any case, Karpathy's idea is more of a concept of what the future may look like.
Absolutely fascinating. This is a new beginning in how to set up custom LLM.
it's interesting, but i don't quite understand the purpose. Is it supposed to make apps easier to write? Hardware easier to add? As unpredictable as LLMs are , it sounds like chaos but maybe I just don't get it.
Couldn't agree more
It's more a concept for what interactions with AI might be like in the future - a fully personalized experience that is responsive to your desires.
@@Leto2ndAtreides oh so it's not an os, more of a framework? I feel like calling it an os might not be the best choice in that case.
Can you imagine building a basic LLM-based OS kernel with just a terminal, where you would use prompts to build every single one of the applications?
That'd be awesome! 😎👍
I don't get the purpose of this either. Everything you have shown can be done with Open WebUI and Ollama too thanks to the recent addition of tools and functions combined with LLMs finetuned for function calling such as the latest Llama3. Storing stuff in a vector database isn't new either, this just gets you an expert system since the actual difficulty is for an LLM based AI system to decide when to pull what in the context and how much while quickly getting rid of trash. Also on conceptual level an LLM shouldn't replace the kernel since like the book indicated this is just an abstraction layer to make the computer actually usable. The LLM should be more or less on the same level as the user. It may need an own memory management system though that could even be driven by the same LLM just like MemGPT demonstrated. So what actually would be helpful would be an LLM that has a good understanding about the operating system and a very direct access to it without trying to replace it. With this it should be able actually do administrative tasks such as detecting errors, checking logs, performing updates and installations even if there are difficulties to overcome. Finally it should be able to get actual work done. I'm wasting so much time with system administration while at the same current LLMs have such a deep understanding that should be able to do most of the tasks on their own.
I think he's doing a basic example of what Karpathy means... An early version of what is to come.
be sure to check out Frdel/agent-zero it's like a less polished, more recursive version using agents that call agents and respond to their creator, even able to troubleshoot itself - to a point
Thank you.
Interesting!
Please explain the ‘Cons’ also 😊
This I wanted to keep more like a how-to guide
No offense but this is presentation is not even close to be an OS. It's just like RAG or adding some tools in your app, so that llm can invoke a particular tool based on user's input, like when to hit a search api.
If this app do task like
"I have deleted my financial report of july, can you recover it from recycle bin?"
or
"can you install xyz game, the file is in the download folder?"
then we could have something different here.
Anyways your content is awesome and very helpful.
Cheers
Definitely no offense taken. I can completely understand that. I also had the initial view like that. That's why for many months I didn't make the video about this. But then I thought it's a good start. You never know so made the video
can not be done locally w ollama?
many local models aren't that good with function calling
Allows ollama intake of openai?
isn't Karpathy's LLM OS all kind of predictable - When did he first mention it?
I think he first envisioned it many, many years ago but very recently. Probably within a year
@@1littlecoder cool and thanks for reply. Would've thought he mentioned much earlier as have seen others moot similar in regard of LMMs going back over a year or 2
so basically its like jarvis from iron man
Do NOT burn your $$$ on OpenAI :) Serious.
Instead use open source llms 😏
Why?
Is this a ollama web ui alternative? That’s what my sleepy brain is gathering
those who said ollama... ollama need systemd which is part of the OS so... it's like which come first chicken or egg LOL 😂
More like OpenAI MLM OS 😂
yea this is crap. you would need a lot more than python to make a os worth using. this is nothing more than anythingllm from what i see.