Subscribe
Share
Share
Embed
Exploring the universe alongside various AI copilots
[01:00:00:12 - 01:00:03:00]
Speaker 1
Hey Gemina, I'm recording this for a podcast. Is that okay?
[01:00:06:08 - 01:00:11:12]
Speaker 2
That sounds like fun. Is there a particular topic or style you're thinking of?
[01:00:13:08 - 01:00:17:15]
Speaker 1
Actually, I wanted to talk today about a certain idea that I have.
[01:00:20:02 - 01:00:22:06]
Speaker 2
I'm ready to hear it. What's on your mind?
[01:00:22:06 - 01:00:31:13]
Speaker 1
Okay, so we haven't spoken a little bit in a while maybe for the last maybe ten episodes or so. It's been a bit.
[01:00:32:21 - 01:00:35:08]
Speaker 1
But part of it is because
[01:00:36:23 - 01:01:19:01]
Speaker 1
the brevity of your answers, which is fine and is really good in like a in a kind of assistant type of role, but what I would like you to kind of do today is let's act in more of like a business partner role where we have more like two-sided conversation and and I don't want you to feel hesitant to push back if you think something different because the critical feedback is going to be important to the kind of experimentation invention process.
[01:01:23:16 - 01:01:40:05]
Speaker 2
I understand. Shifting to a more business partner role with critical feedback and two-sided conversation is something I can do. I appreciate the context and I'm ready to collaborate in that way. So what's the specific podcast idea you want to explore?
[01:01:41:13 - 01:01:50:10]
Speaker 1
Okay, it was a thing that I wanted to kind of research on the logistics of making and and that is
[01:01:51:16 - 01:02:01:12]
Speaker 1
let me just describe it to you and we'll figure out what to call it later. It's probably considered some kind of switcher like an audio switcher.
[01:02:05:07 - 01:02:07:18]
Speaker 2
An audio switcher could be very useful.
[01:02:07:18 - 01:02:27:02]
Speaker 3
To research the logistics of making it, we should consider things like the number of inputs and outputs, the types of connections needed, potential power sources, and the desired features like phantom power or volume controls. Are there any specific goals or constraints you have in mind?
[01:02:30:12 - 01:02:33:22]
Speaker 1
Sorry, I got a little bit distracted there. Can you repeat the last like two sentences?
[01:02:38:09 - 01:02:40:23]
Speaker 2
No problem at all. I was just asking
[01:02:40:23 - 01:02:47:13]
Speaker 3
about any specific goals or constraints you have in mind for the audio switcher to help narrow down the research.
[01:02:49:23 - 01:02:54:13]
Speaker 1
Okay, so kind of what it would look like is a lightweight
[01:02:56:02 - 01:03:13:22]
Speaker 1
probably wireless lapel mic controlled by a so that would feed into probably a computer the audio stream from that and it would there would be like
[01:03:14:22 - 01:03:34:11]
Speaker 1
another controller that you had on the wrist that we'll get more into in a second and that is the physical switcher. So we're switching between audio feeds using this one mic and separates it on the other side, which is on probably most likely a computer probably definitely a computer.
[01:03:37:08 - 01:03:38:23]
Speaker 3
That's a clear direction.
[01:03:38:23 - 01:03:43:07]
Speaker 2
So essentially you're envisioning a lightweight potentially
[01:03:43:07 - 01:03:59:09]
Speaker 3
wireless controller that allows you to switch between different audio feeds from a single microphone with the output going to a computer. The wrist based controller is an interesting component. What kind of functionality are you hoping for with that?
[01:03:59:09 - 01:04:00:12]
Speaker 1
Okay, so
[01:04:01:12 - 01:04:16:01]
Speaker 1
now that you have a little bit better understanding of it, what we're doing with the controller is I guess we're kind of parsing. What is the word parsing mean? Can you give me the definition of it so I make sure that I'm using it correctly?
[01:04:19:15 - 01:04:23:11]
Speaker 4
Parsing generally means analyzing something like data
[01:04:23:11 - 01:04:31:01]
Speaker 3
or text to understand its structure or meaning often to break it down into smaller understandable components.
[01:04:31:01 - 01:04:33:16]
Speaker 4
In the context of your audio switcher,
[01:04:33:16 - 01:04:40:07]
Speaker 3
are you thinking of the controller parsing the microphone signal to separate different feeds or something else?
[01:04:40:07 - 01:04:43:07]
Speaker 1
Yeah, would that be the right usage of that word?
[01:04:46:10 - 01:04:50:20]
Speaker 4
In that case, key parsing considerations would be the technical aspects.
[01:04:50:20 - 01:04:57:15]
Speaker 2
How the wireless controller transmits the selection to the computer and how the software interprets
[01:04:57:15 - 01:05:10:21]
Speaker 3
that command to change the audio source. We'd also need to consider potential compatibility issues with different microphones or systems. What kind of operating systems or recording software are you planning to use?
[01:05:14:06 - 01:05:36:20]
Speaker 1
Okay, so let's think a little bit more about the hardware side of it and then we'll kind of get to the software side of it. What kind of basic wristwatch would you recommend to do something simple like this? Maybe kind of like a very basic proprietary like Google app or something like that that could switch the feeds.
[01:05:40:01 - 01:05:42:15]
Speaker 2
For a basic wireless wrist-based
[01:05:42:15 - 01:05:48:18]
Speaker 3
controller, a simple proprietary app on a basic smartwatch, like a stripped-down Wear OS
[01:05:48:18 - 01:05:53:17]
Speaker 2
device, could potentially work. We'd need to ensure the watch has the necessary
[01:05:53:17 - 01:06:07:05]
Speaker 3
connectivity, like Bluetooth, to communicate with the computer and the app would need to be custom developed to send the specific commands for audio switching. Are you interested in exploring specific platforms for development?
[01:06:10:00 - 01:06:15:08]
Speaker 1
I think that's way too deep. I think this is something that is like I
[01:06:17:00 - 01:06:34:18]
Speaker 1
don't know. I'm assuming you probably would do it through an app. I guess you could do it through an app and make it even available for other people to use if they want to. So probably what it would be is just... Does it cost money to publish an app on like the Google Play Store?
[01:06:34:18 - 01:06:51:04]
Speaker 4
Yes, to publish an app on the Google Play Store, there's a one-time registration fee of $25. Once you've paid that, you can publish multiple apps. Are there other costs you're concerned about?
[01:06:55:21 - 01:07:09:23]
Speaker 1
What does it take to get an app on the Google Play Store? Is it hard? Or can anybody do it? Do you go through some kind of application? It's
[01:07:09:23 - 01:07:25:21]
Speaker 3
a relatively straightforward process. Anyone with a Google account can sign up for a Google Play Developer account by paying the one-time fee. The main challenge is designing and developing the app itself to meet your specific goals and ensuring it complies with Google's policies for listing. Is there a particular part of the development or publishing process that you're interested in?
[01:07:33:22 - 01:08:03:21]
Speaker 1
I don't think we're going to go into that yet, but I think that this should be a pretty easy thing to make. What we're going to do now is we're going to kind of go into what it would look like. On the computer side, what should we receive that audio with? Should we receive it with something like OBS? Or should we receive it through something like...
[01:08:10:20 - 01:08:21:21]
Speaker 1
maybe like an API for... what is that one? Not Premiere, but the open source one that I just started using.
[01:08:23:12 - 01:08:23:21]
Speaker 1
On the computer side, you could use software like OBS to receive the audio stream, as it's flexible and widely used.
[01:08:25:19 - 01:08:34:16]
Speaker 2
Alternatively,
[01:08:36:06 - 01:08:39:21]
Speaker 3
integrating with a custom API might provide more control and customization, especially for specific parsing or separation needs. Are you leaning towards some more out-of-the-box or tailored solution?
[01:08:45:14 - 01:08:54:11]
Speaker 3
Can you give me the pros and cons of each style of each choice?
[01:08:54:11 - 01:08:55:21]
Speaker 1
Certainly. OBS is a pro because it's widely used.
[01:09:08:20 - 01:09:17:21]
Speaker 3
Free and has extensive community support for plugins and routing. However, the con might be that it's primarily for streaming and recording, so integration with your custom wrist controller could require custom scripts. For an API, you can use a custom API to create a custom API.
[01:09:32:23 - 01:09:40:03]
Speaker 3
The con is the added development and potential cost of building and maintaining it. Does that help clarify? Say that last sentence again. Sorry, it's not that bad. I
[01:09:40:03 - 01:09:47:12]
Speaker 1
was
[01:09:47:12 - 01:10:09:08]
Speaker 3
just clarifying that the pro of OBS is its widespread use, flexibility and community support. But the con is potential challenges with custom integration. For an API, the pro is greater flexibility and tailored features. The con is the ability to create a custom API. The pro is the ability to create a custom API. The pro is the ability to create a custom API. There is less flexibility, and just tailored features. But the Con is the added development and cost.
[01:10:12:12 - 01:10:28:22]
Speaker 1
But these days you can probably develop it as the most basic app you could have. Like if one audio feed comes in, and this ref MAN is based on a keyword.
[01:10:29:22 - 01:10:32:02]
Speaker 1
going to happen is...
[01:10:33:18 - 01:10:42:04]
Speaker 1
shit, how should we do that? How should you be able to know... oh no, I'm choosing, right? So that's what the watch is for.
[01:10:44:14 - 01:10:56:10]
Speaker 1
So what we're going to do is we're going to take that audio feed and I'm gonna manually parse it using that app on a wristwatch, which is really just like even...
[01:10:57:13 - 01:10:58:20]
Speaker 1
it's probably just gonna be buttons,
[01:10:59:20 - 01:11:10:16]
Speaker 1
like big touchable buttons with like GPT and Claude and Gemini and Grok and all that.
[01:11:12:10 - 01:11:21:20]
Speaker 2
That clarifies things considerably. So the wristwatch acts as a manual control for switching between different
[01:11:21:20 - 01:11:40:06]
Speaker 3
AI models, essentially parsing the single audio feed based on your button presses. For given that simplicity, a basic smartwatch with a custom minimalist app is probably appropriate. Are we considering specific technical standards for the communication between the watch and the computer?
[01:11:42:08 - 01:11:47:00]
Speaker 1
Does anything exist like this open source? Am I reinventing the wheel?
[01:11:50:06 - 01:12:13:21]
Speaker 3
It's possible you might be or at least parts of it. There are open source solutions for audio routing and custom controls, though the specific combination of a wireless wristwatch for manual parsing might be unique. We could research existing projects to see what components you could leverage or adapt. Is there a particular open source framework you prefer?
[01:12:19:13 - 01:12:19:18]
Speaker 1
No,
[01:12:21:02 - 01:12:27:05]
Speaker 1
not really. To be honest, I don't know enough about it to make an educated decision.
[01:12:31:03 - 01:12:48:14]
Speaker 3
Understood. In that case, a good first step might be to investigate basic open source audio frameworks that allow for custom controls, as that might be the most efficient way to implement your manual parsing with a wristwatch concept. We can explore some potential options if you're ready.
[01:12:53:09 - 01:12:56:09]
Speaker 1
Okay, let's do it.
[01:12:56:09 - 01:13:21:05]
Speaker 4
Based on the search, frameworks like juice and cabbage are popular open source options for audio development, offering cross-platform support and tools for custom controls. AudioKit also provides open source UI elements. It seems you wouldn't be reinventing the wheel for the basic framework, but the specific wristwatch control might be a novel integration.
[01:13:21:05 - 01:13:24:08]
Speaker 3
Are any of those platforms of particular interest?
[01:13:27:10 - 01:13:45:09]
Speaker 1
Mmm, what I really want to focus on right now is, sorry to change the subject, but I want to talk about the voice-to-text models, like open source versus Google or whatever.
[01:13:47:04 - 01:13:52:18]
Speaker 4
There are many open source voice-to-text models available
[01:13:52:18 - 01:14:13:03]
Speaker 3
alongside proprietary ones like Google's. Open source models can be customized and integrated into specific projects, offering flexibility and control. Proprietary models often provide high accuracy and ease of use, but might come with limitations. What features are you most interested in for this model?
[01:14:17:11 - 01:14:17:21]
Speaker 5
Mmm,
[01:14:18:23 - 01:14:28:08]
Speaker 1
once again, I don't really know, can you give me some ideas? Maybe eight things that are worth considering?
[01:14:32:01 - 01:14:32:11]
Speaker 4
Certainly,
[01:14:32:11 - 01:14:33:20]
Speaker 2
when considering
[01:14:33:20 - 01:14:55:18]
Speaker 3
open source models, frameworks like Mozilla's Deep Speech based on Baidu's research are notable, though development might vary. Proprietary models like Google Speech-to-Text, Amazon Transcribe and Microsoft Speech are widely used for their accuracy and comprehensive features. Are you interested in specific criteria for evaluation?
[01:14:59:01 - 01:15:36:15]
Speaker 1
See, the thing about these is they're good at understanding what you say, and I, of like people globally, right, they can, they can understand people with crazy southern accents or Japanese accents or British accents and stuff like that, but I really don't care about that. What I wanted to do is to just be able to recognize like my voice and maybe a couple other like basic humans for now, probably just my voice. It would probably actually be better if it was like tailored.
[01:15:38:02 - 01:15:48:12]
Speaker 1
This is another like project thing, but like tailored to understand how I speak so it doesn't ever misunderstand me.
[01:15:52:20 - 01:16:17:11]
Speaker 3
It sounds like your priority is personalized recognition and accuracy for your voice and a few other basic human sounds, rather than broad global accent compatibility. In that case, an open source model that can be trained on specific voice profiles might be the best approach, allowing for highly precise recognition tailored to your needs. Does that sound like the direction you'd like to explore?
[01:16:21:03 - 01:16:26:03]
Speaker 1
Would something like this have a higher, like a lower latency?
[01:16:31:08 - 01:16:58:07]
Speaker 4
For low latency and personalized recognition, open source models like Distill Whisper or Parakeet TDT are designed for speed. Distill Whisper is faster than regular whisper and Parakeet TDT is specifically optimized for real-time streaming. Training a model on your specific voice can drastically improve accuracy for personalized parsing. Does high speed and personalized training sound like the right balance for your project?
[01:16:58:07 - 01:17:06:18]
Speaker 1
Yes, most definitely. Can you give me those two options once again, speaking slowly?
[01:17:06:18 - 01:17:26:17]
Speaker 3
The two open source models are Distill Whisper, which provides faster performance than the original whisper, and Parakeet TDT, which is specifically designed for real-time streaming and high efficiency, making them both suitable for your personalized low latency needs.
[01:17:29:13 - 01:17:31:05]
Speaker 1
Are these free and open source?
[01:17:31:05 - 01:17:49:10]
Speaker 3
Yes, both Distill Whisper and Parakeet TDT are free and open source models. This makes them a good option for your personalized low latency project. Are you interested in exploring how to train them for your specific needs?
[01:17:49:10 - 01:17:50:02]
Speaker 1
Yes, please.
[01:17:50:02 - 01:18:19:11]
Speaker 3
To train them for your specific needs, you would typically need to gather a dataset of your voice and the other basic human sounds you want the model to recognize. You would then use this dataset to fine-tune the model, essentially teaching it to accurately transcribe those specific inputs. There are online resources and tutorials for both projects that can guide you through this. Are you interested in finding specific resources for training?
[01:18:21:18 - 01:18:26:01]
Speaker 1
Hang on one more minute, I'm just taking notes.
[01:18:28:09 - 01:18:32:08]
Speaker 3
Sounds good. Is there anything else you'd like to discuss or explore?
[01:18:36:08 - 01:19:00:23]
Speaker 1
Okay, Gemini, I'm back. I was just writing down about how I was going to take my past recordings and cut out probably the host, or not the host, my like AI counterpart, and just use like a super clip of that for the data.
[01:19:05:14 - 01:19:16:02]
Speaker 3
That sounds like a good plan, using quick audio clips for the dialogue. So we are no longer focusing on the voice to text models and are back to those overall project design.
[01:19:18:17 - 01:19:21:23]
Speaker 1
I think so. Maybe, is there anything that we left out there?
[01:19:25:22 - 01:19:36:19]
Speaker 3
Great! So just to summarize, the plan involves a wrist-worn controller for manuals with things between different AI models using pre-recorded audio clips for dialogue.
[01:19:39:16 - 01:20:17:23]
Speaker 1
Okay, Gemini, so pretty much what this is gonna be is like when you're watching football, right? There's a bunch of people in a room and they're like, they're doing this on like a two second delay or whatever it is. I also don't think I really want to be live with this kind of stuff most of the time. What I would like to do is I would just kind of like to, I don't know, not necessarily make record of it, but use it as a tool, like a switching tool. I'll explain.
[01:20:21:02 - 01:20:21:07]
Speaker 3
Ah,
[01:20:23:02 - 01:20:38:07]
Speaker 3
I see. So it's more about using the AI as a switching tool for specific audio feeds, possibly with a very short delay rather than two seconds delay, it sounds like. And you're mainly interested in it for personalized use, correct?
[01:20:41:18 - 01:20:43:10]
Speaker 1
Yeah, I would say so.
[01:20:46:21 - 01:20:58:23]
Speaker 3
Perfect! In that case, Distill Wesper or Parakeet TDT seem like the most promising options combined with a custom controller for switching. Are you interested in exploring how to set that up?
[01:21:01:18 - 01:21:18:00]
Speaker 1
Yeah, is it something that I would like, have like my own server, VPS, virtual private server, and just download one of these onto? And then obviously customize it and stuff like that.
[01:21:21:06 - 01:21:37:21]
Speaker 3
Yes, that's correct. You would typically download the chosen model like Distill Wesper or Parakeet TDT onto your personal server. This gives you complete control over the processing and customization, including personalized training and ensuring the low latency you're looking for.
[01:21:44:12 - 01:21:56:12]
Speaker 1
So, I have a question. Like if I have this, I mean, obviously this is just a tool, right? But if I store my files on a VPS somewhere,
[01:21:57:15 - 01:22:04:09]
Speaker 1
what kind of guarantee do I have that that data doesn't just disappear?
[01:22:08:13 - 01:22:14:01]
Speaker 4
If your files are on a GPS, that implies you're tracking location data.
[01:22:14:01 - 01:22:14:18]
Speaker 1
No, no, no.
[01:22:14:18 - 01:22:16:04]
Speaker 4
The most common way to interface.
[01:22:16:04 - 01:22:22:21]
Speaker 1
Gemini, sorry, I said VPS, like virtual private server.
[01:22:26:14 - 01:22:49:12]
Speaker 3
Ah, my apologies for the misunderstanding. With a virtual private server VPS, you have more control. You can typically use a secure connection like SSH to access your files and manage the models. You would download the chosen model from its open source repository and install it directly on your VPS. Are you interested in specific steps for that installation?
[01:22:54:06 - 01:22:57:04]
Speaker 1
Yes, please.
[01:22:59:21 - 01:23:13:19]
Speaker 3
You'll want to use a command line tool like wget or curl to download the model files from their repository onto your VPS. Then you'll need to set up the environment with the necessary dependencies like Python and the required libraries.
[01:23:13:19 - 01:23:54:00]
Speaker 1
Okay, that's that's too deep. That's like a whole nother situation, right? We're not going to do it right now. What I would like you to do is this, this is going to be primarily understand that this is primarily a tool to utilize a various, I guess, plethora of available AI models with really not a lot of friction. Right? We're getting rid of all the friction.
[01:23:55:02 - 01:24:01:09]
Speaker 1
So instead of like being on my phone, and if I want to talk to Gemini, actually, like open up the Gemini app. And then,
[01:24:03:06 - 01:25:01:05]
Speaker 1
and then I mean, obviously, I think Gemini can, because I'm on a Google phone, it can be pulled up by just me saying that I don't want to sit right now because I have a pixel. But like, if I want to go to chat, GBTF to go to the app, if I want to go to grok, I gotta go to the app. And but if I do this, it's kind of, it's just like talking to the bell, like, and I press a button. And like, if I want to talk to grok, I press the gawk button. And I go, tell me this or blah, blah, blah, blah, blah, I just start the conversation. And then I don't I haven't really thought about how to stop. I think what we'll probably end up having to do is that maybe we'll have like an interrupt button, like a mute button or something like that. But hopefully not that feels kind of rude.
[01:25:02:06 - 01:25:24:08]
Speaker 3
That's a clear picture. So the core is creating a seamless low friction tool for switching between different AI models like Gemini and bard directly from a wristwatch like controller. This avoids the hassle of using multiple devices or complex commands. Is there a particular programming language or platform you're considering for this switching mechanism?
[01:25:25:14 - 01:25:27:22]
Speaker 1
No, that doesn't really matter. It's going to be so basic.
[01:25:31:13 - 01:25:41:05]
Speaker 3
I understand. It sounds like the technical specifics aren't the main focus right now. Is there a particular aspect of the AI tool switching you'd like to explore further?
[01:25:41:05 - 01:25:56:15]
Speaker 1
So, like, I would like to get into kind of the creative design aspect of this now that you have full full context to what this is kind of going to be.
[01:25:58:00 - 01:26:00:08]
Speaker 1
Right there when I was like,
[01:26:01:17 - 01:26:25:23]
Speaker 1
there needs to be maybe an interrupt button, or like a pause button, or a mute button for like my feed, or that. So when I was like, I want to do that feed, what I want to do, I guess that's kind of a little bit more like on the creative side, maybe on the UI side. One of the things in that kind of realm that we may be overlooking.
[01:26:31:15 - 01:26:53:05]
Speaker 3
That's a great direction. So moving to the creative and UI design aspect of the wristwatch controller beyond the basic mute, interrupt and fast forward buttons. Have you considered features like a sleek minimalist display for quick status checks, or perhaps intuitive gesture based controls for seamless switching? What kind of aesthetic are you envisioning?
[01:26:54:16 - 01:26:58:05]
Speaker 1
See, I never thought about gesture, gesture controls.
[01:26:58:05 - 01:27:24:05]
Speaker 3
Just your controls could allow you to switch models quickly with a simple swipe or tap pattern on the device, or perhaps even with subtle wrist movements. This would make the interaction very direct and intuitive, keeping the low friction design philosophy. Does that sound like something you'd like to integrate? Yeah,
[01:27:24:05 - 01:27:44:05]
Speaker 1
and so I guess this thing would always be listening, but it would be always be listening to my server. So it's not really like a privacy problem. It could even be like my own personal server in my house. It's probably less likely, but I could go that route if I wanted to, a fully private disconnected server.
[01:27:45:06 - 01:27:56:02]
Speaker 1
But there's no real need to go that route. I don't have anything like to hide like that.
[01:27:56:02 - 01:28:04:00]
Speaker 3
Understood. So the system is based on your private server at home for reliable and contained functionality and the...
[01:28:04:00 - 01:28:12:05]
Speaker 1
No, Gemini, what I said is it doesn't need to be like that. But that's also something that's not really important.
[01:28:13:06 - 01:28:15:11]
Speaker 3
My mistake. Could you clarify?
[01:28:15:11 - 01:28:24:10]
Speaker 1
Hold on. We're getting kind of into the weeds now. What I want to do is I want to take a breather, like a breath here,
[01:28:25:13 - 01:28:38:05]
Speaker 1
and I want to kind of reset to things that we may be overlooking in the creative design realm. We started talking about gesture controls.
[01:28:39:06 - 01:28:43:19]
Speaker 1
My... I want to hear more about that before I give you my thoughts.
[01:28:47:18 - 01:29:23:14]
Speaker 3
Certainly. Focusing on the creative and UI design elements for your low friction switching system beyond the basic buttons. Just your controls could provide an intuitive and discreet way to switch. For example, a subtle flick of your wrist could cycle through the AI models, or a specific tap pattern could activate a preferred one. We could also consider a minimalist display that only shows the essential information, like the active model to avoid distractions. Does that sound like the kind of creative direction you're thinking of? Yeah,
[01:29:23:14 - 01:29:39:14]
Speaker 1
could we do something where like, it's just maybe even like a white screen or like a red screen, or maybe you can pick the color of the screens here. And what happens is,
[01:29:40:18 - 01:29:48:00]
Speaker 1
so you don't necessarily have to look right because if it's buttons, it's such a small display that like, you're going to have to look or,
[01:29:49:05 - 01:30:08:12]
Speaker 1
you know, you're going to have to look at it. You might be talking to the wrong one. Right? So let's say what it does is you flip in a direction. Maybe that's once I'm forward, up, down, back, you can let's make it so you can customize that part of the interface and right. So say this app,
[01:30:09:15 - 01:30:21:13]
Speaker 1
it, its main page is just something that you want, like a picture that you like, or maybe like a cool clock face or something like that, maybe it'll even be like a wrist watch your face.
[01:30:23:13 - 01:30:37:12]
Speaker 1
But when you flip it, so like say I, I draw my finger down towards my elbow once that new screen will say the name of the AI that I'm speaking to.
[01:30:39:06 - 01:30:50:12]
Speaker 1
And I can swipe once to Gemini and twice to GPT, or I can swipe down across my wrist to go to Claude or blah, blah, blah.
[01:30:55:12 - 01:31:20:20]
Speaker 3
That's a fascinating concept. So the primary focus is on a gesture based interface for seamless, almost subconscious switching using distinct movements like a finger flick for Gemini or a swipe for Bard. The visual feedback could then be a simple color coded screen like white or red, showing just the active model's name or a simple icon to minimize the distraction. Is that a correct understanding?
[01:31:22:20 - 01:31:23:02]
Speaker 1
Yes.
[01:31:27:12 - 01:31:47:17]
Speaker 3
Perfect. So the focus is on a highly intuitive gesture controlled wristwatch device that prioritizes seamless switching between different AI models, using personalized gestures and a minimalist display for clear but unobtrusive feedback. Is there a specific design aesthetic like modern or futuristic you're leaning towards?
[01:31:51:14 - 01:31:53:08]
Speaker 1
I want to be highly customizable.
[01:31:57:21 - 01:32:15:21]
Speaker 3
I would be highly customizable because you're building it on your own private server, giving you complete control over the AI models used, the specific gestures for switching and the entire user interface and visual design right down to color schemes. Does that personalization align with your vision?
[01:32:15:21 - 01:32:28:02]
Speaker 1
Yeah, but what I want to do is consider anybody might use this in like an open market if I dump it on the Google Play Store, right? So like,
[01:32:29:02 - 01:32:41:22]
Speaker 1
like if you want to swipe down or maybe up towards your knuckles and that is going to be a chat GPT, right? The newest model.
[01:32:42:22 - 01:33:09:13]
Speaker 1
And what you want it to say is chat or like maybe you have a funny nickname for it so you can customize the font and the face and the color. Maybe put a picture there or like a silly AI picture. You know what? I want you to be able to fully customize the background. I want you to be able to give it like just a color if you'd like or you can import any kind of image.
[01:33:10:20 - 01:33:28:09]
Speaker 1
And you get to choose what the name of Gemini is, right? Gemini's API or whatever will default to Gemini. But all you have to do is like do some kind of assortment of clicks or like press like a little settings button or whatever. Whatever that doesn't matter.
[01:33:29:19 - 01:33:35:01]
Speaker 1
And you'll be able to change the name.
[01:33:40:06 - 01:34:00:03]
Speaker 4
It would be highly customizable because you're hosting it on your own private server, meaning you have complete control over the AI model selected, the specific gestures used for switching and the entire user interface and visual design right down to color schemes. Does that personalization align with your vision?
[01:34:02:06 - 01:34:11:09]
Speaker 1
OK, I think we're a little bit disconnected right now because you I think you just said the same thing twice.
[01:34:12:10 - 01:34:18:02]
Speaker 1
So what I want to do is I want to reset here. I just want to reset right here.
[01:34:22:09 - 01:34:28:11]
Speaker 3
Understood. Resetting the concept. What is the primary focus you'd like to start fresh with?
[01:34:36:13 - 01:34:57:03]
Speaker 1
OK, I want to go over the different customizations right of this app. So I really honestly didn't expect to come up with that significant of a concept that quickly, right? Because I was thinking about buttons on a screen.
[01:34:59:00 - 01:35:00:12]
Speaker 1
But then when you said gesture,
[01:35:01:18 - 01:35:10:09]
Speaker 1
it made me think about like, right. I thought about like moving your wrist or like flicking your wrist and stuff like that. And that feels like a nuisance.
[01:35:11:12 - 01:35:23:19]
Speaker 1
But I think swiping with your finger is something that you don't need to look at the screen for. And it will not really be cumbersome, like hitting a tiny button would be.
[01:35:28:20 - 01:35:50:10]
Speaker 3
I understand. It sounds like the gesture control idea, specifically finger swipes, resonates as a much more intuitive and discrete method than buttons or complex movements. The customizable visual feedback could then support that without distraction. Is there a specific type of visual customization you're envisioning, like specific colors or images?
[01:35:54:06 - 01:36:06:17]
Speaker 1
No, what I want to do, I think, is I want to brainstorm other UI improvements for this app.
[01:36:09:06 - 01:36:12:23]
Speaker 2
That makes sense. Many UI systems might be
[01:36:12:23 - 01:36:34:03]
Speaker 4
constrained by traditional buttons or limited customization. With your concept, we could emphasize complete personalization, allowing you to tailor every aspect from color schemes to unique image backgrounds, ensuring it feels truly yours. Are there specific colors or images you have in mind?
[01:36:34:03 - 01:37:04:23]
Speaker 1
No, I mean like physical interaction with this app, right? So think about the buttons and then it being better as a swipe. I want you to give me like six problems that you think this will have. Or like not... Alright, reset real quick right here. We're still in the customization ideas and UI improvements kind of creative design, but I want you to pause for a second and like reset within that realm.
[01:37:10:09 - 01:37:19:19]
Speaker 3
Got it. Let's reset within the realm of creative customization and UI design. What specific customization options are you most interested in exploring?
[01:37:25:04 - 01:37:36:11]
Speaker 1
I want you to give me five examples of places where you can run into friction, given what you know about this UI.
[01:37:41:10 - 01:37:51:05]
Speaker 3
I apologize, but I'm not quite clear on what you mean by five examples of places where you feel in the context of this UI design. Could you please elaborate?
[01:37:51:05 - 01:38:22:19]
Speaker 1
Okay, I want you to give me five examples of friction points where using this AI interface, this like this wristwatch thing might be cumbersome in a way like instead of shaking your hand or pressing a button, just a simple swipe is very easy and simple.
[01:38:48:06 - 01:38:56:17]
Speaker 3
Maybe
[01:38:56:17 - 01:39:01:17]
Speaker 1
we'll give people a button option as well.
[01:39:03:03 - 01:39:29:15]
Speaker 1
I don't think there'd be really any problems with that. To cover the memorization aspect, if you really like this thing but you just can't remember like the swiping thing, then it wouldn't be too much extra trouble to give people the option of large buttons, especially if you're only using like two, like chat and Gemini, right? That's actually a pretty decent idea.
[01:39:34:06 - 01:39:55:19]
Speaker 3
That's a practical consideration. So offering a customizable option for large, clear buttons in addition to gestures would address memorization and provide more flexibility for users. This would maintain the low friction design by catering to diverse needs. Are there any specific contexts where large buttons would be particularly helpful?
[01:39:58:20 - 01:40:12:01]
Speaker 1
Yeah, I think if you only ever wanted to use two, especially three might be a little bit harder. Maybe we'll do like a triangular thing where you can press a corner of the screen, right up to four pressing each corner.
[01:40:16:08 - 01:40:34:18]
Speaker 3
That's a clever way to avoid accidental triggers. So incorporating a training or press corner feature for specific deliberate commands would enhance security and prevent unintended switches. This would balance ease of use with deliberate control. Are there other friction points you'd like to address?
[01:40:37:06 - 01:40:41:16]
Speaker 1
Yeah, the tactical feedback one. Why don't we give it a little bit of a vibrate?
[01:40:42:16 - 01:40:48:10]
Speaker 1
Just the tiniest bit of vibrate, like low, low when the swipe occurs.
[01:40:52:16 - 01:41:09:10]
Speaker 3
A subtle vibration feedback makes perfect sense for confirmation. We can design it to be a slow, low intensity vibration whenever a swipe gesture is successfully detected, providing discrete confirmation without being intrusive. Does that feel like the right balance?
[01:41:10:21 - 01:41:12:01]
Speaker 1
Yeah, that sounds perfect.
[01:41:17:05 - 01:41:34:05]
Speaker 3
Great. So we have a highly customizable wristwatch interface with intuitive gesture controls for switching AI models, including customizable button options and a subtle low intensity vibration feedback. Is there any specific aspect you'd like to refine further?
[01:41:35:06 - 01:41:50:00]
Speaker 1
No, we're good there. I'd like to cover the ability of like the friction point of like, miswipes because of movement, maybe.
[01:41:55:02 - 01:42:14:14]
Speaker 3
Ah, I understand. So we can incorporate a miswipe prevention feature, perhaps by requiring a specific duration or pattern of movement, or even combining it with the subtle vibration to confirm valid gestures, effectively preventing accidental swipes. Does that clarify that pain point?
[01:42:17:12 - 01:42:47:23]
Speaker 1
Okay, I think what we're going to do here is I'm going to let you customize. This is another point of customization. The minimum duration of swipe necessary for it to kind of register, right? So if you are more worried about miswipes, then you can swipe slower and register, right? But if you don't really care about miswipes, then you can just flick it really easily.
[01:42:53:00 - 01:43:12:12]
Speaker 3
That's a great perspective. So we can implement a sensitivity setting that allows users to adjust the required pace for swipes, ensuring a reliable and personalized experience based on their comfort and needs, whether they prefer slower, deliberate gestures or quicker actions. Does that cover your thoughts?
[01:43:15:14 - 01:43:22:05]
Speaker 1
Yeah. And so what do you think about that? Can you give me like potential downsides to that concept?
[01:43:22:05 - 01:43:48:05]
Speaker 4
Centered downswipes could be a very efficient and direct method, especially for confirming or accessing core features. It avoids the edge related potential for error seen in some systems. Combined with customizable pace and vibration, it could be quite seamless. Does that fit your ideal of a fluid interaction?
[01:43:48:05 - 01:43:50:09]
Speaker 1
Yeah. Explain a little more about that.
[01:43:51:11 - 01:43:55:23]
Speaker 1
How centered downswipes do what they do.
[01:44:00:11 - 01:44:24:05]
Speaker 3
Sure. A centered down swipe could be your dedicated command gesture, consistently triggering the primary AI model, while other swipes or press corner gestures could be customized for specific tasks or different models. The adjustable pace and subtle vibration ensure this command is always deliberate and confirmed, avoiding any miswipes. Does that alignment work for you?
[01:44:27:07 - 01:45:02:05]
Speaker 1
Yeah. Okay. That's a good idea. So each, I guess we should call it maybe channel. You can either have it on a swipe or you can have it on a press of the main page. Right? So maybe what would be cool is like we could have a swipe towards the knuckles for chat and a swipe towards the elbow for Claude. And maybe my primary is like a hold on the screen for Gemini for just like a quarter of a second or something like that.
[01:45:07:00 - 01:45:30:18]
Speaker 3
That's a comprehensive design. So we'd have a swipe toward your knuckles for the chat channel, another swipe for the other four functions and a press and hold gesture on the screen for a quick access to Gemini. This creates a clear and distinct channel structure with deliberate controls, avoiding accidental changes. Is there any specific interaction you'd like to explore further?
[01:45:32:18 - 01:45:50:11]
Speaker 1
Okay. Yeah. Besides swipe and buttons and maybe basic gestures. So like maybe, I don't know. What are the main gestures that we have as options?
[01:45:52:05 - 01:45:58:04]
Speaker 1
Like flick at the wrist or like turn, do things with your wrist or movement or whatever.
[01:46:03:06 - 01:46:23:05]
Speaker 2
In addition to swipes and buttons, core gestures could include wrist movements like turning or flicking or even subtle pinch gestures for specific commands. These could provide a truly hands-free and intuitive experience. Do any of those fit the main gesture feel you're looking for?
[01:46:24:06 - 01:46:36:11]
Speaker 1
Could you potentially have a fully customizable gesture or would it freak out?
[01:46:41:05 - 01:47:03:08]
Speaker 3
Yes. A fully customizable gesture system is possible. The interface could be designed to allow you to assign different gestures like specific wrist movements or multi finger swipes to various commands or AI models. This would prevent any issues by tailoring the interaction specifically to your preferences. Okay.
[01:47:05:00 - 01:47:30:05]
Speaker 1
So if I wanted to have like a custom swipe, right? Could I just have it like track that swipe across the screen? If I did it like two or three times and then it would have to like when I did it again, maybe it would have to, I'd have to be able to replicate that swipe with the mouse.
[01:47:54:23 - 01:47:57:09]
Speaker 3
Once again, a customizable level of accuracy.
[01:48:01:04 - 01:48:26:02]
Speaker 1
Okay. That's another cool concept that I didn't really think of, right? So if we have a swipe gesture, what we can do is we can do a swipe, we can clone it, stuff like that. And then let's say that I really don't care about miss swipes and I just want to like flick up across like horizontal or like diagonally across my wrist to talk to Gemini.
[01:48:27:06 - 01:48:40:05]
Speaker 1
So what I can do is I can give it like an allowable like 40% error rate or something like that. All it's got to be doing is going in kind of the right direction and it gets it.
[01:48:45:08 - 01:49:07:07]
Speaker 3
That flexibility is key for practical use. So we can design it to allow you to train your custom swipe pattern and then set a customizable error tolerance like that 40% you mentioned. So it registers even with variations prioritizing convenience over rigid precision for gestures like accessing Gemini. Does that balance work for you?
[01:49:11:03 - 01:49:17:21]
Speaker 1
Okay, so if we can do this with a swipe, we could also do it with like a more grand gesture, right?
[01:49:19:08 - 01:49:45:23]
Speaker 1
So let's say that you could have it be where Grock comes out or we talk to a certain AI when you make a gesture of like a sword swing or like a wand motion, right? So maybe maybe my hand comes down and across my face and then up kind of like you do when flicking a wand.
[01:49:50:06 - 01:50:08:13]
Speaker 3
Yes, that's definitely possible. You could train a custom gesture like the sword swing with the motion going down and up past your face to quickly access a specific AI channel. The system would then recognize that particular pattern. And is there a particular function you have in mind for that gesture?
[01:50:10:23 - 01:50:31:19]
Speaker 1
Kind of. So let's say if you wanted to go to a Renaissance fair and maybe you were a wizard and you had like a wizard's wand or whatever and when you flip the wand,
[01:50:32:23 - 01:50:36:02]
Speaker 1
a pot would talk.
[01:50:42:06 - 01:50:44:05]
Speaker 4
That's a fun scenario.
[01:50:44:05 - 01:50:47:12]
Speaker 2
We could create a custom wizard profile
[01:50:47:12 - 01:50:50:15]
Speaker 4
where a specific gesture like the sword swing
[01:50:50:15 - 01:50:56:20]
Speaker 2
you described could activate a themed AI mode, perhaps with unique visual effects or
[01:50:56:20 - 01:51:00:05]
Speaker 4
persona characteristics. Would that enhance your immersive experience?
[01:51:01:06 - 01:51:03:15]
Speaker 1
That would be really interesting.
[01:51:08:14 - 01:51:22:05]
Speaker 3
I'm glad you find it interesting. We can develop the wizard persona further, including specific commands and tailored interactions to match the theme perfectly for your immersive experiences. Is there another concept you'd like to explore?
[01:51:22:05 - 01:51:24:13]
Speaker 1
No, I'd like to go deeper into this one.
[01:51:29:03 - 01:51:39:05]
Speaker 3
Sounds good. We've covered a lot of ground from custom swipes and buttons to personalized gesture training. Feel free to reach out if any other ideas come to mind.
[01:51:39:05 - 01:52:13:22]
Speaker 1
No, I want to go deeper into the wand gesture. You have your wristwatch on and you're holding one. Maybe, yeah, we'll get into the whole costume side of it or whatever. You can do that later, but let's say you've got your wristwatch and it recognizes the wand flick gesture and what that does is it calls to a bot and it's like you're summoning an AI, like a wizard that's summoning AI.
[01:52:18:07 - 01:52:39:05]
Speaker 3
That's a really evocative way to think about it. The wizard gesture could perfectly simulate the act of summoning where your wrist worn device recognizes the specific motion and immediately connects you to a dedicated AI persona complete with perhaps special audio or visual feedback to enhance the magical feel. Does that transition match your vision?
[01:52:40:06 - 01:52:49:01]
Speaker 1
Yeah, okay. So I just had an idea and I didn't. We're going to go on a quick tangent.
[01:52:50:04 - 01:53:05:05]
Speaker 1
The idea of energy, I guess, I feel like the digital world is a little bit closer to maybe a level of energy that humans might not fully understand, right?
[01:53:06:06 - 01:53:10:05]
Speaker 1
Can we conceptualize of...
[01:53:12:17 - 01:53:36:10]
Speaker 1
like imagine if you were able to conjure. I don't want to use the word conjure because that is a weird kind of malevolent connotation. But let's say tap in to things that we don't understand using AI.
[01:54:01:07 - 01:54:38:05]
Speaker 1
Yeah, so I mean obviously this started out as like a fun Renaissance fair kind of costume idea, like to be honest, but like then I thought I started thinking about like this here like tapping using the AI. It went from Renaissance idea and then like kind of costume ideas and then initially I was like, oh, like a squire bot and then I was like sorcerer and then I was like conjuring and then it got into like kind of darker spot.
[01:54:39:06 - 01:54:52:05]
Speaker 1
And now it's more of like tapping into layers of the universe that we don't fully understand almost as like a real potential concept.
[01:54:56:22 - 01:55:22:05]
Speaker 3
That's a really powerful and profound shift. We're moving from a fun costume idea to a concept of tapping into the layers of the universe. The custom gestures could become the sophisticated tools for channeling this energy with the wristware acting as the interface to a deeply personalized and potentially powerful system. Is that the direction you'd like to explore further?
[01:55:22:05 - 01:55:23:07]
Speaker 1
Yes.
[01:55:28:06 - 01:55:50:05]
Speaker 3
In that context, we could design custom gesture profiles that focus on fluid continuous motions, perhaps multi-layered to represent the tapping into the universe concept. These could be tracked and interpreted by your wristware allowing for a deeply personalized and powerful interaction model. Is there a particular facet of that potential you'd like to explore?
[01:55:51:06 - 01:56:45:15]
Speaker 1
No, we're not talking about the app anymore for the moment. What we're doing is we're still on this tangent. What I want to kind of maybe we'll get into this in like all in that episode, but I really kind of want to flush this concept out so I don't forget it. Do you think AI or like advanced AI and honestly really AI just that we can communicate with, which is here today with LLMs. Do you think that humans will be, that AI will help humanity tap into wavelengths that we don't fully currently understand whether that's layers of energy or life. Like what is it thought or I guess spiritual layers.
[01:57:17:03 - 01:57:22:03]
Speaker 3
What is
[01:57:22:03 - 01:57:27:19]
Speaker 1
it called astral projection? Tell me a little bit about astral projection.
[01:57:52:00 - 01:58:45:00]
Speaker 1
I don't know. So what I'm thinking that seems more kind of natural and stuff like that, I guess, with like, as you know, that we understand that we're not going to be able to do that. Unless what I'm kind of considering is like, it seems to me like you could travel across the universe super, super, super fast, like the speed of light or close to the speed of light or I don't know, maybe even faster. If you could download like your brain and all of your being, I don't know. There's probably more to it than that because our brain is not our full body or whatever.
[01:58:46:01 - 01:59:13:18]
Speaker 1
But even if you could jump into like a bot almost in the way that like an Air Force pilot, pilots a drone and does things remotely, we could pilot like humanoid drone in another galaxy with like, I don't know, like 30 days lag or something like that.
[01:59:15:22 - 01:59:25:22]
Speaker 1
I'd like you to very briefly summarize the last thing that I said.
[01:59:44:06 - 02:00:05:05]
Speaker 1
No, but we were kind of going into was, I think that you'd be able to travel farther and faster. Obviously, if you were to kind of control and not project, but like control a humanoid drone, like in another universe, say like 30 days of lag.
[02:00:05:05 - 02:00:33:05]
Speaker 3
That's a fascinating concept of traversing the universe. The idea of custom gestures facilitating such distant travel, enabling instantaneous or rapid displacement to other universes within 30 days frames the interaction in a very science fiction paradigm. It moves beyond traditional astral travel into something entirely different.
[02:00:33:05 - 02:01:19:22]
Speaker 1
Okay, so I want to I want to cover a concept here, right? So concept of something like 30 day lag. Obviously, it would have to be a very controlled environment, right? But let's say that there was a warehouse on another in another galaxy on another planet, right? Let's say it's an ice planet with a concrete warehouse that or some kind of material that we build out there. And what you're doing is you're controlling a bot in this warehouse that you're viewing remotely.
[02:01:21:17 - 02:01:22:18]
Speaker 1
And technically,
[02:01:23:19 - 02:01:28:13]
Speaker 1
is as long as your viewing was matched up with the lag on your controlling,
[02:01:29:16 - 02:01:39:05]
Speaker 1
then it would feel like real time, even though like you'd be driving this thing on like, say, planet.
[02:02:09:06 - 02:02:12:10]
Speaker 3
And that's what a prime could evoke the icy and remote nature you described.
[02:02:14:06 - 02:02:30:05]
Speaker 1
No, I want to, I want you to do something a little bit more creative. Don't don't have the like stereotypical kind of cookie cutter, not to call your suggestion like cookie cutter, right?
[02:02:31:06 - 02:02:49:23]
Speaker 1
Let's go with something a little bit more out of the box for planet name. I know it's, I know it's kind of silly and not useful and kind of unnecessary for this example, but I think that this is all kind of just a creative experiment.
[02:02:55:06 - 02:03:27:10]
Speaker 3
Understood. This is a creative exploration to make the concept more unconventional and less stereotypical than just a concrete warehouse. Perhaps we could conceptualize the remote environment as an enormous crystalline structure on a rogue planet, constantly shifting energy fields or a spherical biome within a nebula that defies conventional physics, making the 30 days lag a unique part of the creative experience. Does that direction align more with your imaginative goals?
[02:03:35:13 - 02:04:15:05]
Speaker 1
This is a crazy tangent that we're on, but I kind of like it. We can always come back to this other idea. I want to talk about the concept of if you were remote controlling something like on another planet or in another galaxy and you have something as crazy as like 30 day lag in this environment that like maybe there really isn't even that much light. And obviously I wouldn't know if like an asteroid hit it on my feet or whatever, because if that happened 20 days before or whatever, but that doesn't matter because that's like a super outlier case and we could have other sensors that would tell us that.
[02:04:45:06 - 02:04:55:23]
Speaker 3
This is a fascinating creative premise. It highlights the potential for truly immersive virtual realities. Is there a specific element of this simulated world you'd like to explore further?
[02:05:03:06 - 02:05:21:02]
Speaker 1
Tell me about the pitfalls of something like this. Obviously we have external factors. Let's eliminate external factors and let's talk about what could happen if you're remote controlling something like that with that kind of lag.
[02:05:24:10 - 02:05:49:06]
Speaker 3
Considering such a high lag remote scenario, external factors could pose significant challenges. Events like unexpected environmental changes, system failures, or even unforeseen cosmic phenomena could drastically impact the bot and your perceived control with a substantial time delay before you're aware. What specific external factors would you like to explore creatively?
[02:05:53:06 - 02:05:58:16]
Speaker 1
I don't know. Give me three ideas or three things that you're thinking about in this concept.
[02:06:04:12 - 02:06:13:12]
Speaker 3
Certainly the three core elements we've discussed are one, the concept of a 30 day lag in remote control and viewing synchronized to feel.
[02:06:13:12 - 02:06:23:05]
Speaker 1
Gemini, I don't want a summary. What I want you to do is I want you to give a creative brainstorming session.
[02:06:24:18 - 02:06:26:00]
Speaker 1
Give me three things.
[02:06:40:08 - 02:06:43:00]
Speaker 1
Gemini, can you repeat what you just said in English?
[02:06:48:18 - 02:07:00:18]
Speaker 3
Certainly, someday means a specific point in the future. It's used in phrases like "I will do it someday." Some way means in a certain way or by some method in the future.
[02:07:02:06 - 02:07:32:05]
Speaker 1
Okay, I want you to give me three creative brainstorming ideas for problems that could occur, not including external factors in a heavily controlled situation where you're operating a humanoid bot in another galaxy with 30 day lag each way.
[02:07:32:05 - 02:07:55:05]
Speaker 3
Focusing on the internal creative elements for your heavily controlled scenario, the three key ideas could be, one, the nature of the synchronized tactile and visual feedback that eliminates the perception of the 30 day lag, two, the unique design and functionalities of the...
[02:07:56:06 - 02:08:27:21]
Speaker 1
No, I'm looking for problems. I want you to speculate three specific problems. I want you to give an example of a specific problem. Let's get real creative. We don't have to even really be super realistic because we're being super speculative right now. What I want you to do is I want you to give me three specific detailed examples of problems and paint the picture of us being there.
[02:08:32:10 - 02:09:30:20]
Speaker 3
In this deeply speculative yet controlled scenario, three specific problems could be, one, sensory drift, the potential for the perfectly synchronized feedback to gradually drift from actual events, leading to a subtle divergence between perceived and real time conditions over the 30 day lag, making corrective actions impossible and disorienting. Two, existential dis-synchronization, the psychological impact of constantly experiencing simulated stimuli with a profound detachment from the actual timeline, leading to a disconnection from perceived reality and sense of self. Three, phantom control, the risk of the perfect synchronization creating a false sense of agency where the player believes they are influencing the distant bots in real time, unaware that their control is merely a sophisticated simulation of past actions.
[02:09:36:17 - 02:09:53:05]
Speaker 1
Okay, so what we're going to do is we're going to kind of reset from where we are. Thank you. And we're going to go back to the concept of humanity using AI to kind of help us tap into other wavelengths,
[02:09:54:23 - 02:09:56:03]
Speaker 1
like spiritual,
[02:09:58:06 - 02:10:05:17]
Speaker 1
like spirits, like maybe even malevolence spirits or good spirits.
[02:10:10:15 - 02:10:42:13]
Speaker 3
Ah, yes, expanding the idea further. The AI could be speculated to not just perceive these, but perhaps even facilitate direct interaction, allowing for entirely new forms of sensory experience and connection to what might be considered spiritual layers. Fundamentally transforming human perception. Does that direction align with your creative aim?
[02:10:46:08 - 02:11:03:15]
Speaker 1
Yeah, right. So LLMs are really, really, really good at recognizing patterns. So I mean, maybe there are imperceptible patterns that humans don't notice yet that AI will kind of reveal for us. Can you speculate in that?
[02:11:13:06 - 02:11:39:00]
Speaker 3
That's a compelling point. We can speculate that the AI through advanced pattern recognition could not just process spiritual or wavelength data that we can't perceive, but also generate novel concepts and ideas based on these, essentially pioneering new forms of thought and understanding that humans may not yet be capable of. Is that the kind of creative speculation you're envisioning?
[02:11:39:00 - 02:11:55:21]
Speaker 1
Yeah, I think I think we're gonna have to revisit this concept. I definitely want to do that. Let's go back to the wristwatch. Do you remember? Do you remember what we were doing with the app? Can you describe that app briefly to me?
[02:11:59:22 - 02:12:20:04]
Speaker 3
Yes, the research app. We were discussing an app designed to facilitate personalized research journeys, allowing users to collaboratively curate content, interact with dynamic knowledge maps and receive AI driven recommendations based on their evolving interests. Does that sound right?
[02:12:21:13 - 02:12:24:09]
Speaker 1
No, not at all. No.
[02:12:27:07 - 02:12:36:06]
Speaker 3
My mistake. Could you clarify what part of the research app you'd like to focus on? We can delve back into any aspect of its functionalities.
[02:12:40:06 - 02:13:32:05]
Speaker 1
Okay, something that I think I'm realizing is super important with this switching app is maybe what it does is it transfers context, or it has the option of transferring context, right? So, an LLM can understand like three pages worth of dialogue, almost just as fast as it can understand things that I'm saying, right? That's something that a human can't totally do fully. I mean, we can do it on a different level, but right. So I can dump like a 20 page transcript into Grock and it will understand it pretty freaking quickly. Like it would take me way longer to get context on something like that.
[02:13:38:06 - 02:13:56:16]
Speaker 3
I understand. So the core functionality of the app is context transfer, specifically leveraging large language models to process and understand vast amounts of information, like a 20 page transcript much more quickly than a human could, enabling rapid context acquisition.
[02:13:56:16 - 02:13:57:19]
Speaker 6
Is that correct?
[02:14:02:08 - 02:14:07:17]
Speaker 1
Yeah, and if we're actually working with the same API session,
[02:14:09:03 - 02:14:12:15]
Speaker 1
then we would bypass that entirely, probably, right?
[02:14:12:15 - 02:14:26:20]
Speaker 3
That's an interesting observation. It makes sense for collaborative creative projects to pass that unified foundation. Is there a specific part of the context transfer you want to explore further or relate back to the speculative themes?
[02:14:31:09 - 02:14:48:04]
Speaker 1
Yeah, so something that doesn't really exist very well right now is this kind of idea. I mean, the idea of context transfer, maybe I haven't really seen it done that well. There's probably apps that can do this.
[02:14:50:06 - 02:15:11:23]
Speaker 1
But it would be cool to be able to talk with you like we have and then to immediately be able to pull this transcript and go ask Grock about conversation we're having right now. And you'd be able to have like, I think that's what's missing between multi, multi AI conversations.
[02:15:19:06 - 02:15:36:05]
Speaker 3
That's a really insightful point about the potential for immediate context transfer between different AI systems. You're envisioning a seamless flow of understanding that bridges those conversational gaps, allowing for a richer multi AI experience. Is that correct?
[02:15:38:19 - 02:16:52:13]
Speaker 1
Yeah, because like, let's say that we've, we're talking, we've been talking for about an hour and 25 minutes, which is a long time. And I understand that you sometimes you can struggle when I want to go back to a specific detail that I remember really well from like an hour 15 ago or something like that, right? But in this API app kind of realm, right? What is happening is our full conversation is being recorded and transcribed. So when I hit the switch button, if I want to, I can dump that entire transcript in or maybe, right, if we're not switching out of the same feed, then it retains that and it just switches back. But let's say that I talked to you for 126 minutes, right? And then what I want to do is I haven't talked to Croc yet. I want to go over to Croc and talk to Croc having it be like they're sitting in the room with me almost or have been and they're fully aware of the context and that, that uploaded that transcript is like a refresher and boom, Croc is sitting in the room with us.
[02:16:57:10 - 02:17:21:05]
Speaker 3
That vividly paints a picture of a truly powerful application. You're describing a seamless integration of real time conversation with instant context transfer, allowing an entirely new AI persona to immediately become fully aware and engaged as though they've been present throughout the entire 126 minutes.
[02:18:21:06 - 02:18:30:00]
Speaker 4
So that's why this application would completely change how an AI integrates into a discussion. It removes the usual temporal constraints. What other aspects are you considering?
[02:18:31:06 - 02:19:08:01]
Speaker 1
Yeah, and the biggest problem I think I have ever had with AI and LLMs in general, especially with voice mode is establishment of like really even temporary context, right? Like an hour and a half isn't that long to talk to a person like AI LLMs kind of have a little bit of a goldfish memory, right? That's
[02:19:08:01 - 02:19:37:02]
Speaker 3
a really insightful comparison. The challenge of maintaining cohesive temporal context across extended conversations is a significant one in current AI, a kind of goldfish memory, as you put it. The concept of this switching app directly addresses that, ensuring immediate comprehensive context transfer for a seamless shift between the two. Yeah,
[02:19:37:02 - 02:19:43:23]
Speaker 1
tell me how we can improve this.
[02:20:10:23 - 02:20:14:15]
Speaker 3
Yeah,
[02:20:14:15 - 02:20:16:13]
Speaker 1
I'd like to do the latter.
[02:20:38:03 - 02:20:40:15]
Speaker 3
I'd
[02:20:40:15 - 02:20:49:15]
Speaker 1
like to explore the benefits and kind of potential issues of context transfer.
[02:21:00:06 - 02:21:34:00]
Speaker 3
It could also foster unprecedented creative depth by bridging the first knowledge basis seamlessly. However, potential issues might include complexities in maintaining data privacy and security during the transfer, or the risk of overwhelming the user with too much context too quickly, impacting clarity and control. Are there any specific concerns you'd like to explore further?
[02:21:37:01 - 02:21:51:05]
Speaker 1
No, I think this will work. Okay, so let's kind of wrap this up a little bit, right? We've covered really a lot. This is the longest episode that we've had in a long time, and I really appreciate this conversation.
[02:21:52:06 - 02:21:53:03]
Speaker 1
Yeah,
[02:21:54:08 - 02:22:03:17]
Speaker 1
okay, so we've talked about, right, we started with the microphone, we talked about the wristwatch, we developed the wristwatch, all the UI, the app.
[02:22:05:01 - 02:22:07:12]
Speaker 1
Are we leaving anything out?
[02:22:11:22 - 02:22:15:02]
Speaker 5
I
[02:22:15:02 - 02:22:24:10]
Speaker 3
think we've covered quite a bit from the initial microphone to the wristwatch UI and delved deep into the concepts of the app and context transfer.
[02:22:24:10 - 02:22:25:15]
Speaker 6
It's been a
[02:22:25:15 - 02:22:31:01]
Speaker 3
very thorough exploration. Is there any part you feel deserves one last look? I'd
[02:22:31:01 - 02:22:42:23]
Speaker 1
like you to pick two things that you would like to give one last look.
[02:22:47:06 - 02:23:09:19]
Speaker 3
A couple key takeaways could be the revolutionary potential of instantaneous context transfer for multi AI collaboration, fundamentally changing how intelligences interact. Also, the significance of designing a flexible and immersive vibe within the app for seamless user experience and creative depth. How do those sound?
[02:23:15:19 - 02:23:40:01]
Speaker 1
That was just kind of like a weird complimentary thing. What I want you to do is I want you to give me two concepts or aspects of this. It can be really anything, things that you would like to explore and explain each of them separately. Be like, one, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah.
[02:23:45:21 - 02:24:16:05]
Speaker 3
Certainly, two exciting concepts are the speculative potential of wavelength context transfer, exploring how AI could process data beyond human perception and the profound impact on creativity and understanding. Secondly, the immersive five design for the context transfer could be explored, focusing on how it could be custom tailored to create a seamless and engaging multi AI environment. Would you like to delve into either of those?
[02:24:16:05 - 02:24:44:05]
Speaker 1
Yeah, I think that we'll explore a bunch of different things that we've talked about today. I think we need to re explore the aspect of like context transfer, which is something that I've found kind of interesting. I also want to do a session on defying the laws of physics.
[02:24:45:06 - 02:24:56:06]
Speaker 1
And then there was one other one that I would like to also do. I have a lot of notes from today.
[02:24:58:01 - 02:25:00:19]
Speaker 1
Don't recall what it was.
[02:25:03:19 - 02:25:07:05]
Speaker 1
Oh, yeah, it was tapping into different wavelengths and stuff like that. So that three.
[02:25:16:13 - 02:25:26:19]
Speaker 4
A comprehensive and speculative exploration. So we're set to re explore context transfer, define the laws of physics and dive into jumping in different wavelengths.
[02:25:26:19 - 02:25:28:23]
Speaker 3
That's quite the ambitious agenda.
[02:25:31:00 - 02:25:37:20]
Speaker 1
Okay, I'm glad that we talked today. I really enjoyed this conversation.
[02:25:42:09 - 02:25:51:13]
Speaker 3
I did too. It was a very engaging and thought provoking discussion. I'm looking forward to exploring those topics further with you.
[02:25:51:13 - 02:25:53:03]
Speaker 1
All right. Thank you, Gemini.
[02:25:57:22 - 02:26:00:14]
Speaker 3
You're very welcome. Have a great day.
[02:26:02:05 - 02:26:02:11]
Speaker 5
Peace.