Certainly possible :) I didn't think an "edible Wi-Fi router" would resonate as well lol which was the original "new idea" I was gonna use so I picked this one out of the many responses as I thought it was relatively unique and illustrated the idea of combining two things pretty well.

This is a great point actually - the models do tend towards the average which I think is an interesting phenomenon and I wonder if it can be surpassed given our current architectures and training approaches. Writing effective copy is something I've had a hard time getting ChatGPT to do well. Unique individuals with unique experiences will have unique neural connections and thus can generate more innovative ideas (in other words, less statistically likely ones). I think it's possible to replicate this in AI but it might not be easy!

But this is a *desk* 😉 haha but yes it's hard to come up with anything entirely new I suppose. It did also tell me "Edible Wifi Router" maybe I should've used that idea!

These are great additional considerations! People definitely have unique perspectives which helps avoid the constantly "average" response that AI often provides. I think this can be partially imitated by telling the AI which perspective it should take (e.g. "you are a marketing expert"). This is the approach used in Agent Swarm systems where multiple AI Agents are working together to solve a problem. Discovery is an interesting subject, similar to Understanding in my mind - there is relativity and nuance there. Current AI Agents do have the capability to identify that they don't have all the data they need to answer a question, at which point they can invoke a function call to retrieve data, usually via web or knowledge base search. But I don't think this is what you mean - you are referring to bigger picture discovery of new things that are not yet represented in the data the AI can access. This is indeed much harder, because the AI would need to take all the steps you described (plan, experiment, observe, conclude) which is difficult to set up with current systems. I think we'll be able to run certain experiments like this if we have good simulation environments with APIs on top but that's a big complex task we haven't explored yet. So definitely a lot of work to do on this front before AI can just start generating new laws of physics, medicines, etc. It will be interesting to see how these systems evolve. Right now they're able to use all that data to help us brainstorm and combine ideas. Next, we'll have agents that can coordinate a bunch of steps and take action to get things done. After that, perhaps we will see enhanced abilities to execute experiments in a controlled environment. Most interesting part to me is that it's all primarily a data integration problem now instead of a hypothetical "will this ever be possible" sort of problem.

Thank you for the kind words :) Yeah I kept pondering and discussing this with a few folks and just decided to organize my thoughts and make this. Really helped clarify my own thinking!

Thank you for watching and for your kind comment :) Glad you enjoyed the video!

Hey Denson :) Thanks for supporting the channel since the beginning! Would be glad to connect - you can find my contact info here: www.youtube.com/@VoloBuilds/about

Glad you enjoyed it, thanks for watching! I've been thinking about these subjects a lot lately. Trying to find some good content to evolve my thinking about the concept of emergent properties. Vsauce has been interesting. Keep thinking, my friend! ... but also don't think too hard and just go with the flow 😉

Haha yes, if we consider combinations of things as "not new", this is another way to put it! But I would argue that things can/do emerge as we combine them, so we could call them "new" - bricks arranged into a house creates a new house (which wasn't there before) - though I suppose we could debate that too (I actually just watched "Chairs Don't Exist" by Vsauce and it was very interesting lol)

You have a keen eye! 🫡

How about "Edible Wi-Fi Router"? That was one of the others

Lots of cool creative AI stuff rising up everywhere! I think my points apply across all of these systems