Thanks for bringing this to my attention Sam. I watched their "Take Control of the World" demo video. I liked that the agent "understood" that it needed to be in compliance with intergalactic law before expanding outside Earth. I have respect for that type of due diligence lol
This video, as all your videos are, is such an amazing resource. I can't thank you enough for taking the time to walk through what is happening and why. I would love to see this expanded upon and perhaps have the agents able to access a Pinecone Vector store where they can reference a premade "Library" of documentation to help them in their conversation. I really appreciate all the effort you put into these videos.
great content, really love it. I'm thinking about implement it as an business consultant, asking more and more questions untill a second agent says,"ok, enough gathering, time for solutions" .
The name choice for inception prompting is attributed to a concept expressed in the film "Inception", as noted at the foot of page 3 of the paper: 3.1 Role-playing Framework “What’s the most resilient parasite? An Idea. A single idea from the human mind can build cities. An idea can transform the world and rewrite all the rules. Which is why I have to steal it.” - Dom Cobb, Inception
Things are moving at the speed of Moore's Law squared. I get into one project and then another juicy thing comes along. However, I discovered a solution: Create my own juicy thing. I got a few ideas...
would it be possible to use GPT4All instead of openai models? I have tried to make something like that (I am by far not the most skilled developer…) but I was not able to obtain a decent result.
I think there’s the possibility of an agent doing ToT (train of thought) + an agent solely focused on critiquing which would definitely be able to solve reasoning tasks and or increase quality prompt enhancement/inception.
You basically described an unreleased yet Google Gemini Ai platform. The idea of numerous Ai is not new, in USSR they've made a space shuttle operated by trio computers, self landed successfully. Of course Google will have much more powerful artificial entities.
Hi Sam, I keep coming back to your videos as they're a great resource, and explained in plain english too. So thanks mate. Is there a way hook these agents up to chroma at all? I can't find an example anywhere and would love to be able to give them some context.
Yes these can be hooked up. I will show that at some point, I am working with things like this and I want to make a deeper dive at some point. Also the Minecraft Voyager paper does this so I want to show that at some point.
The "role flipping" problem highlights the most basic weakness of the autoregressive model: the model is completely unaware of which parts of the input/output were produced by whom. It is similarly typical for a LLM to accuse me of having said something which in fact has come straight from the LLM's imagination itself. I bet the LLM has the whole time an impression that it is just talking to itself (why wouldn't it) and no real awareness of any external world or events.
Thanks Sam for wonderful video. Have you got any video on tabular data generation? Medical Billing example on langchain site is very basic example. Looking for a little complex like nested fields or referential integrity etc
Thanks Sam for making such a complicated topic and paper easy to understand. Quick questions: you mentioned few times using synthetic data from LLM to “market research” could you point to some urls where you saw that happening?
I have been involved in helping with bots for companies where it does market research and tasks like generate customer main objections etc. From the feedback I have been told these were very similar to responses from real people.
Few questions: 1. Could this be done over a document concept so sharing pros vs cons based on a user 2. Is memory is included under the hood for this type of chain?
Yes you could get multi agents reviewing a doc etc. You could incorporate memory, depends on what you want to use the memory for as to how best to do that.
In this chapter I like what you are mention that 2 agents can interact with each other. The inception prompting play a key role if I not make a mistake? Is it possible to dive deeper in this, to make it more understandable for my as a beginner? Many thanks.
can you please tell me or make a new video about making a tool or a transoformer agent that can take an audio and dubb it to another language with whisper or Nllb-200 and make a talking avatar to say it with sadtalker for free . thank you very much .
Haha I've done this before, always ends up with the AI agents discussing AI about themselves, quickly agreeing with each other and complimenting themselves about how important AI is and how great their "ideas" about AI are, in a loop.
A local LLM version of this would be incredibly helpful!
Thanks for bringing this to my attention Sam. I watched their "Take Control of the World" demo video. I liked that the agent "understood" that it needed to be in compliance with intergalactic law before expanding outside Earth. I have respect for that type of due diligence lol
This video, as all your videos are, is such an amazing resource. I can't thank you enough for taking the time to walk through what is happening and why. I would love to see this expanded upon and perhaps have the agents able to access a Pinecone Vector store where they can reference a premade "Library" of documentation to help them in their conversation. I really appreciate all the effort you put into these videos.
great content, really love it. I'm thinking about implement it as an business consultant, asking more and more questions untill a second agent says,"ok, enough gathering, time for solutions" .
The name choice for inception prompting is attributed to a concept expressed in the film "Inception", as noted at the foot of page 3 of the paper:
3.1 Role-playing Framework
“What’s the most resilient parasite? An Idea. A single idea from the human mind can build cities. An
idea can transform the world and rewrite all the rules. Which is why I have to steal it.”
- Dom Cobb, Inception
Things are moving at the speed of Moore's Law squared. I get into one project and then another juicy thing comes along. However, I discovered a solution: Create my own juicy thing. I got a few ideas...
Thanks another great video👍 yes an example with an open source model would be Nice. Guanaco llm seems to perform well...
would it be possible to use GPT4All instead of openai models? I have tried to make something like that (I am by far not the most skilled developer…) but I was not able to obtain a decent result.
I think there’s the possibility of an agent doing ToT (train of thought) + an agent solely focused on critiquing which would definitely be able to solve reasoning tasks and or increase quality prompt enhancement/inception.
You basically described an unreleased yet Google Gemini Ai platform. The idea of numerous Ai is not new, in USSR they've made a space shuttle operated by trio computers, self landed successfully. Of course Google will have much more powerful artificial entities.
Hi Sam, I keep coming back to your videos as they're a great resource, and explained in plain english too. So thanks mate. Is there a way hook these agents up to chroma at all? I can't find an example anywhere and would love to be able to give them some context.
Yes these can be hooked up. I will show that at some point, I am working with things like this and I want to make a deeper dive at some point. Also the Minecraft Voyager paper does this so I want to show that at some point.
@@samwitteveenai can't wait :)
The "role flipping" problem highlights the most basic weakness of the autoregressive model: the model is completely unaware of which parts of the input/output were produced by whom. It is similarly typical for a LLM to accuse me of having said something which in fact has come straight from the LLM's imagination itself. I bet the LLM has the whole time an impression that it is just talking to itself (why wouldn't it) and no real awareness of any external world or events.
Thanks Sam for wonderful video. Have you got any video on tabular data generation? Medical Billing example on langchain site is very basic example. Looking for a little complex like nested fields or referential integrity etc
Thanks Sam for making such a complicated topic and paper easy to understand. Quick questions: you mentioned few times using synthetic data from LLM to “market research” could you point to some urls where you saw that happening?
I have been involved in helping with bots for companies where it does market research and tasks like generate customer main objections etc. From the feedback I have been told these were very similar to responses from real people.
This is amazing andI have so many ideas for this. I would love to see what Camel + LangChain + hfAgents could do.
Few questions:
1. Could this be done over a document concept so sharing pros vs cons based on a user
2. Is memory is included under the hood for this type of chain?
Yes you could get multi agents reviewing a doc etc. You could incorporate memory, depends on what you want to use the memory for as to how best to do that.
Thank you Sam for this great video
Magic stuff, as always. Thanks!
In this chapter I like what you are mention that 2 agents can interact with each other. The inception prompting play a key role if I not make a mistake?
Is it possible to dive deeper in this, to make it more understandable for my as a beginner? Many thanks.
I am planning some content purely on prompting so I will cover that in there.
@@samwitteveenai I appreciate it, not only for me but for all the beginners in this space.
Thanks for the awesome video! I actually use agents with different personas in the Design Thinking process, and the results are pretty promising!
Yes I have been experimenting with different forms of roleplay and personas it is a great way to influence the output and can be funny at times.
Great video! keep up the good work!
can you please tell me or make a new video about making a tool or a transoformer agent that can take an audio and dubb it to another language with whisper or Nllb-200 and make a talking avatar to say it with sadtalker for free . thank you very much .
I would love to have the same topic but done with Local LLM's thanks in advance :)
Haha I've done this before, always ends up with the AI agents discussing AI about themselves, quickly agreeing with each other and complimenting themselves about how important AI is and how great their "ideas" about AI are, in a loop.
A local LLM version of this would be incredibly helpful!