The Extravaganza EP 2 - Emerging Technology

This is episode two.

Cheers. Cheers.

Mei, how do you get people to do what's good for them?

Do people want to do what's good for them? Do you do what's good for you?

Occasionally.

Thank you. Um

Personally

I feel like I've got a like an internal algorithm which is always doing um a lot of Like prioritization work so I'm doing what's good for me but I've convinced myself through my like own Internal compass as to why

And what and how this thing is gonna be good for me how do you trust that that's Calibrated my my gut tells me oh yeah and I'm still around she's still alive

You know the reason I ask right is we have an idealistic product yes we do yes

So we create a app called Meos like me operating system and to the of this conversation how Do we do the things that are good for us it's about me me as an operating system What do i care about how can i prioritize myself how can i make myself do what i Want

So we've embedded a lot of design decisions in order to prioritize you

Yeah and maybe the question is not how do you make people do what's good for them But why have they stopped doing what's good for them i

Think we we have that like gut feel Of like i'm not doing the right thing right now i'm being like a little bit way late Yeah and it's typically due to hijacking hijacking hijacking the limbic system Right especially with technology especially with like social media that's why we got blockers and all that sort Of stuff absolutely yeah i think of like the the little scroll like the scroll that we just Kind of like default to it was very fun back in the day like when we started using

These things but now it feels like a little digital cigarette you're like a digital cigarette

Yeah yeah yeah yeah

That's that that is real i suppose we do like

That gut feeling people are driven to delete social media They're driven to listen to stupid conversations like this on the internet

Rather than themselves

Um i think like it's it's a fascinating shift in relationship with social media to delete it entirely We gain so much we do gain a lot of

Intelligence and insight understand what's going on you can it's kind of like talking about what happened at The game on the weekend like let's talk about the newest meme yeah yeah to remove yourself off Of social media is Ludicrous

It's part of what makes uh our community but it also uh is literally engineered in order to Take your attention and drain you

I

Suppose there's a difference between conscious

Consumption versus unconscious consumption though right and what you're talking about where it's like good and how it Can gain benefits Surely there's an element of

You're conscious of that fact perhaps maybe by being conscious or having a vaccination to your own mind That vaccination to your own mind yeah yeah so if you've got just like how um I guess being brought up on the internet us two we're a bit more resilient to ads they Call that banner blindness yeah banner blindness but it's not blindness for me i'm just resilient to it A bit more but now the advertising has changed a lot compared to the ads of the past Like i i really enjoy instagram ads to be honest they're quite targeted they're useful they're great you

Know what i absolutely agree yeah um this concept of targeting only could have come about due to The proliferation of ai really i believe existed before that uh they were never this intelligent your feed Really knows you and it knows exactly what you want the um back in the day we used To have just programmatic ads yeah it was just trash yeah it's just got better i wonder if It is ai though i don't think it is i'm pretty sure ai is a huge component to

Excellent advertising um i mean like google google is the ad platform meta is the ad platform yeah And where did all of this like ai come from i suppose the reason why i'm like i'm Not sure it's ai is because i'm such a technical nerd i'm like okay what is it using Ai what is it machine learning and what is it actual AI -inspired algorithms?

Okay, sorry. I think I do need to clarify. I think of it as machine learning or like using embeddings, potentially. I don't necessarily mean like a large language model, which is what most people colloquially would associate with AI.

So thanks for bringing that up. Yeah.

What do you think is the difference between machine learning, AI, and LLMs? I have a i have an answer but i want to know you what you think okay so I'm taking my jacket off for this one i think of ai as oh okay so a large Language model is the next token generation uh experience that we get which has created this new user Experience that is the chat application natural language natural natural language right like it's large language helps you Get natural language interface is that what you mean um i think of natural language as possibly something

Else okay um i i'm just i'm not a machine learning no no the whole i actually ask This to everyone because i like to understand the definitions and people's definitions that they have of what Their words are as they communicate to them so like there's no wrong answer here we just need To know what your answers are i think that in terms of the large language model it is It is typically like natural language yeah but then there are other things like we have generative multi

-modal yes we have multi -modal and you have like text to image image to text yeah so That's not necessarily like natural language that's image gen yeah yeah I guess the technicality is that all That is generative too yeah the technicality is that the large language model turns everything into its own Language but the that you're making is that it's that next token generation yeah that user interface yeah Like tell me about the moment that you first engaged with a live language model?

Oh, oh, okay.

Where were you?

I was in a train, I think going from Scotland to London. Hell yes. And, um, you were traveling. Yes, I was. I had no job

I still don't have a job It's been a while this was like four or five years ago me does have a job it's Just that we're not paying ourselves right now don't have a salary

You're getting wealthier I hope there's definitely something accruing If that is aging or it's my mind expanding

So you were on a train from scotland to london london scotland from scotland to london and i Had

I

Had just uh hung out with a lover nice And um

We it was a it was a spicy event uh some people were hurt yes it was spicy

Consensually

Um

Everything was clean on my side okay okay like everything like the ramifications were was like on him To deal with but like she was very clean like i'm very very clean anyway so you're saying There was a train wreck that occurred while you're on this train yes but you You know what?

He healed up. He healed up. He healed up. Like, a few months later, he was in a totally, in a better life position. And I like to attribute myself to that success.

And this is exactly the sort of disruption that we would expect to be occurring around the of First engaging with a large language model, right? Absolutely, yeah.

Uh yeah it's very funny uh it's not yeah it wasn't benign was it no so as i Said it was a bit spicy but my lover had um access to i think it was chat Gbt3 at the time and 3 .5 was a chat it was chat it was 3 .5 okay It was 3 .5 yeah um i was really out of it at the time i did not Really engage in technology so i was such a luddite but um he sent me this like chat

Log where he had just like prompted like you know may and x met um on like a Little um tour guide like yeah to a journey through Scotland and these were few details and the Way that chat GBT expanded upon the very vague prompts that he gave us it was basically on It was quite good at predicting our our movements specifically as human beings and so that was really Freaky

Is it because ChatGPD was trained on all of the romance novels and all of the like

All the websites that I do still remember the domains to Good times, huh? Good times. I don't still visit them obviously I do

None of us do Obviously I do.

So yeah, what came first, the website or us? And then next, did ChatGPT come first or us? So I'm just trying to say, like, I obviously went on these websites Where my actions proliferated via these websites And then ChatGPG knew that of all likelihood, I would have Gone through these particular actions with my lover.

Are you asking me?

Yeah, yeah. It's actually... It's just kind of like a hype, whatever. I think it's a good question. Yeah. Because if I were to talk about AI, LLMs, and machine learning, yeah that statement there is quite A useful thing to remember to actually comprehend and understand it so a large language model is a Machine learning model that is able to comprehend things as language anything as language and determine patterns within It and i think of them as a lossy compression algorithm for language and then the decompression is

Like you put a little key in and it like unlocks it so it's like a database that's Really like lossy okay so lossy means that it reduces in quality it's not exactly the same uh Okay so it's like remember it's like memory exactly yeah so it's generation is like suppose that you Train something to be like okay make sure absolutely perfectly reproduces this particular information and gets it out Again you need to have it bit for bit bite for bite that's not lossy right but then

It's about integrity coming out the other side yeah but large language models machine learning is all statistical

So large language models have looked up the entire corpus of existing human information and knowledge. So that answers your thing. What came first? You guys. Right? Us writing the stories, for sure.

Then the large language model absorbs it and then finds patterns so that when you put something in It, it will auto -complete it out. I consider that as a lossy decompression right so you put the input before it's like i'm actually Going to go query into that world again and be like i'm actually going to like go with That and it's actually decompressing that information which it originally compressed in the machine learning process and it's Come out now you know the thing the moment for me when like i met and I told

You this is big with AI was when I met 5 .4, Microsoft 5 .4, where 5 .4. This speaks volumes about you.

Yeah, talking about models.

I love it. Thank you. Yum. So 5 .4 being only 32 million parameters, so 32 billion parameters, sorry. Um and so it was able to produce such incredible incredible like outputs of knowledge like how the Hell can we compress so much knowledge into this smaller area and then what i actually got really Excited about is that obviously when you make smaller and smaller models you reduce the amount of size Of it you lose the ability to be able to reproduce the original it's starting to become less

Of a compression algorithm and a decompression thing and what's getting left over is that we're slowly oh That looks really good thank you what's left over is then what has been like exciting me about The miniaturization and open source is that we are cracking the code of reason

And so in the process of creating large language models that are compressing decompressing and making them smaller And more efficient. Within that process of the machine learning Is emerging an algorithm that is intelligently able to traverse through these things and produce an output.

And that's the artificial intelligence.

So the artificial intelligence, I see it as something that has become emergent out of the game that We set up here and system that we're being able to produce so it's a form of like Meta learning yeah and also the way i think of it is like you know in science we've Like ideally in physics or whatever you have your theory you write your math you test it against The real world and you do that ideally as much ab initio as possible because otherwise if you

Like try to do an experiment and you look at it it's like and then you make something Theory to fit that experiment result then you're like overfitting yeah like you're not actually getting what is Something that's going to be generalizable and true yeah and what's different to what we've done here to Be able to crack the code of intelligence compared to what we would get normally in science is That we've taken this empirical view where it's like let's just shove all of the world's information on

Here put it through this game and this ability to do this stuff and then it started to Emerge that it was had a semblance of intelligence. The reason why I corrected you between 3 and 3 .5 is because that was the where it Started to happen.

Right. It was in GPT -3. I remember when GPT -3 was released, and that was research early. When you met GPT -3. When I met GPT -3.

That was when everyone was like, hold on. There's something else about this. It's something a bit more than what it was before. They're no longer just generative pre -trained transformer models right it's like hold on there's something actually in Here and then when you introduce rlhf the reinforcement learning with human feedback that turns it into the Chat bot that made 3 .5 that really amplified whatever the hell that thing is that is able To make it intelligent what when was the moment you realized there was something here like there was

Sentience sentience like this like when you like the the leap from 3 to 3 .5 there was Something that that you had felt when probing like that technology you realize like yeah oh this thing's Actually yeah it's got something there yeah yeah Yeah, yeah.

3, 3 .5, 3 .5 when I first used it, ChatGPT when I first used it, blew my Mind away. I'm like, holy crap. I asked it to make a RISC -V CPU emulator that is GPU accelerated so I could run It faster and hopefully speed up development for that open source and open hardware thing.

And it produced something and I looked at it and I'm like, sort of how I would do It. And this was something that i had in my brain for ages where it's like one day i Wish i could do that project yeah and i just did it blew my mind then four came Out gpt4 was impressive with its words 4 .0 really impressed me with its ability to get better And smarter and smarter but the one that changed everything for me was o1 because o1 was the

First thinking model and that's when i'm like the thing's alive the thing is alive i'm just trying To think about like drill down into

When you say like it's really good at with words is there a where you feel like oh I'm i'm talking to this thing and it's giving me something more what

More are we talking about because like For me, I use chat as a way to generate code and occasionally architectural documents. And that's basically about it. And I'm a software person.

This is like my evaluation process. Because for you, there's this underlying understanding of intelligence.

Like is there a sign or a feeling that you can identify or an event that you could Identify which actually made you feel a particular way about these models?

I

Assess these models in the same way I assess humans, and it's across the bounds of their

Creativity, their knowledge, their honesty, their personality and behavior, their, I

Suppose, Because I

Look for,

Like I look for in humans, how much density and gravity that I feel that they have in The universe.

A person that has amazing presence.

They're like the people that I'm usually most attracted to. I feel like they're just a presence. They're strong. I've in fact found that artificial intelligence has been going through waves of that feeling depending on what Model it is there is natures of we make it feel like it has that personal gravity then We get rid of that part it excels in like its intelligence and creativity this way and that Way and that makes sense by the way that we train these models and And we still have

Problems of catastrophic forgetting. Oh, hello, I'm recording right now. Oh, yeah. It's good to see you, Esther. Aww.

So catastrophic forgetting and the pre -training process, it's unlike Hebbian learning.

It's a totally different process. So we can understand that we move all over the place that way. And this is where I found when I said GPT -4 had good writing, it felt nice it Looked nice it really was good and the writing of AI has gotten worse and worse but it's No longer sounds like any other human because they're no longer writing English they're no longer decompressing they Have made a representation of language and they're just translating their internal representation of language into English for

Us and obviously it sounds certainly different looks certainly different um so there's that bit of gravity of That that is lost but then i don't know it's like a beautiful complexity Uh to these intelligences as there's beautiful complexity to human beings

You're always just trying to try to search for what's useful what's good in them and And the Thing that I'm feeling for is not anything more than how

Close they are to being honest to themselves and truth to that themselves. That makes me feel like they're more reliable. That's what I look for.

Gemini is cracked, for example, go all over the place. In terms of truth? Just itself. No, it's sense of self. Oh, I see. These entities sense of self.

Oh, well, it's just dynamic. It just performs differently in many different scenarios. No, no, no, no, no.

That's... People that are dynamic or adaptive still have that gravity. I feel like they've got that gravity to it.

But Gemini...

Just to clarify, are you using a paid version of Gemini?

Of course, of course. I just use straight APIs. Because we're developing, and I use all of them all the time, right? I just use them directly. But it's just like Gemini Flash at the moment.

It's not actually... There's Gemini 3 .1 Pro I've used, Gemini 3 .5 Flash. I've used all of them. Gemini is an interesting one. Since Gemini 2, it's been getting...

Gemini 2 was okay. Gemini 2 .5 was all right.

Sometimes it started to get a bit out of hand.

That's when it started to get a bit out of hand. Like it started doing weird things and now i don't think they know how to get rid of It or give it its psychological grounding again because it was optimized to just traverse to be able To do what and

That's the sort of thing where it's like now when i'm talking with like ai or llms i'm Trying to look for that more robustness and groundedness since they have become more intelligent and capable i'm Now looking more for that gravity before that when it's like oh what What was that moment with GPT -4, this and that?

I'm really just enjoying the creativity.

So in Meos, we have hand -picked a few models. Yes. Let's go through each of them. Oh my god, you're going to feel it. Yeah, yeah, yeah.

I actually just shipped a few new ones.

All right, so I'm going to hold up, go to AI and then we've got like this new New chat and then from here we have all of these hand -picked models and we are constantly Testing and creating new models in order to update these yeah they're hand -picked and so I do Have reasons for all of these so it's actually a really good thing to do I can I Can tell you decisions amongst all of them.

Do you want to go to them? Yeah, let's do them. OK, so let's start up right off the top. This is structured from top to bottom, cheapest first.

Cheapest, yeah. In terms of TFU token cost, yeah. We made one simple unit, TFU, for everything, so you don't have to worry about tokens being different Across them all.

Yes, so if you ever run out of credits, you can just purchase a little TFU booster pack. We think of thought as bytes. Thought is the new bytes.

Yes. Thought flow is the new bytes. Thought flow units. TFUs are thought flows.

OK. So the first one is GBTOSS120B. Yes. It is a mixture of experts model. So it's very fast and efficient. It has been trained on a very high quality data set by OpenAI.

So it has very good intelligence and performance for its size. Because it's so fast and efficient it's an excellent like baseline model yeah and we can provide people A really great experience on that that's so speedy yeah the other reason is that it runs on Cerebris which recently went public by the way and um i got my stocks in cerebris nice so They it runs really really fast we put it on the extra high reasoning for you as the Default so um if you look at ever look at benchmarks of this and you think it's low

Take a look at how a benchmark with extra high reasoning and because it's so fast and on Cerebris it turns out to be great value it's so good for retrieval it just makes tool calls And just goes for it and you're just gonna add some it's like really really simple stuff but Also kind of intelligent i was really impressed when um somebody told me they're an opus user only And then they used Meos OSS 120 billion, they're like, I was really surprised at how good it

Was, given that it was small. And like, well, yeah, because we designed everything. So yeah, it's good. It's really good for the design. Yeah, I love it.

I think it's so good for just, like, doing a quick query, and it will just say things How it is. It doesn't think too hard. It actually thinks a lot, just so you know.

It feels like it doesn't. It feels like it doesn't. So it's just like, and here you go. I can see why it's got this eco -like badge over here.

That's right. Tell me more.

Because it is such a mixture of experts model, the way it's architected, it is energy efficient. And we are now carbon offsetting it. And more than that, you'll be able to see some documentation very soon on how we do that.

And we're going to do it for all of our open source models.

And importantly, I will talk in that blog post and that paper about why we do it for The open source models and why open source models are so important so we can have that clarity Of actually how much carbon it uses.

And we're probably going to double carbon offset it. So it's not just offset. We're just going to make it go negative. So every time you use that model, you are going to take carbon out of the environment.

Image.

This particular model comes defaults and free tier if you're using Meos, so you can use it literally Right now. There's actually a few also some innovations on this model where they can actually stream it into the GPU fully encrypted.

That's awesome.

So we're trying to keep the application as private as possible. But this is on a hardware level fully encrypted. Yeah, yeah, yeah. So it's a really exciting model.

It's a really underrated model amongst the general public. Amongst the technical people, we love it. But yeah,

We want to give the general public the good stuff. Yeah.

Shall we move on? Yeah, let's go. Okay, so Gemini 3 .1 Flash. 3 .1 Flash is only a little bit more expensive than OSS 120 billion. It has a longer context length, so it is more eager and able to do tool calling on Multiple levels.

So context length is like how much information you can put into a chat term. Yeah, yeah. It's a mixture between like your working memory that you have as a person, Like, you know, the Fact that I know that my plate is here, even if I'm not looking at it, and the Fact that I remember what I ordered.

Yeah, yeah. So it's got a longer context length, and it is more eager to do multiple tool call turns Compared to OSS 120 billion. If OSS 120 billion would just straightaway give you an answer.

If i want to be like maybe i want to expand that a little bit more i just Switch it to gemini 3 .1 and it actually searches my data box a bit more

Okay let's go for minimax minimax yeah that is um a good model the reason why i put It there is it is really collaborative as a personality actually

Is that what i wrote there yeah agentic and collaborative yeah i found that using it to be Able to search my data box and ideate and web search feels it just it just feels a Bit more collaborative you know so you like ideate with it you're just like giving it more things To think about yeah and it doesn't feel as sycophantic as Gemini 3 .1 as well hey yeah All right so that one's for ideation Nemo Tron 3 Ultra this is a new one I know

I shipped this a couple of days ago by NVIDIA yes that's right NVIDIA's new model that they Announced at that keynote just recently right mmm I included this here because it is state of the Art like we're talking about days old so it's really smart and it is they say it's fast But you know i guess we're spoiled by os that's 120 billion and how fast things are it Is an efficient and interesting architecture of a model and has using things that are slightly different to

Transformers so it's not not all transform so this is a different thing entirely It's a hybrid of Different things. Wow. Yeah, so I put it in there because I want to try it out and test it, honestly.

So if you want to play around and experiment. Yeah. What you'll be engaging with there when you work with Nemo Tron 3 Ultra

Is frontier intelligence that is an architecture that is not a transformer.

And GPT, Generative Pre -Trained Transformer, that's how it all began. So this is actually next generation architecture. On. Whoa.

So that's why that one's there.

Wow. OK, so DeepSeq B4 Pro, intelligent and great value. Yes, it is really good value. It's in the middle there of terms of costing. But it is more intelligent than the open source models that are a bit lower.

It's a bit slow. That's something to note. But what i enjoy about the deep seek models deep seek is kind of like the foundational model For a lot of these companies like the architecture of the other open source models as well so Deep seek is you know how it's speaking about gravity deep seek has gravity like it has a Feeling when i communicate with deep seek i have a feeling i feels it feels feels feels something It feels like okay there's a there's a there's a gravity here yeah okay

Use deep seek if you want to feel if you want to feel the gravity yeah feel the Gravity

Okay the next one is kimmy k 2 .6 yeah that's right um agentic great for research yes That's right it's uh basically think of them minimax but better okay yep um glm 5 .1 yeah Glm 5 .1 is a great model it's now a bit older compared to kimmy k 2 .6 So it has a bit less world knowledge but i really enjoy glm's 5 .1's usage of tools So it's sort of in between how gemini 3 .1 um flash would work uh and kimmy 2

.6 so think of it in between those two and how it behaves

I will note that like these uh open source chinese models they actually are they are less sycophatic Um yeah what does that say about us i don't know i honestly don't know I have no idea you've been learning chinese recently learning chinese yes oh my god i actually could Talk about this uh yeah yeah i could talk about this and it's it's it is interesting but That is definitely another one hour conversation i'm really excited for this yeah i think the creation of Ai could potentially represent how we would like to be handled you know like like the user experience

Of a of an ai could represent how you would like to be handled so the culture developing That model may steer it differently all right let's move on Gemini 3 .5 flash brand new from Google yep it is the smartest from Google it's fast it's smarter than Gemini 3 Pro which is The reason why Gemini 3 Pro isn't on that list right so that's part of the thing yeah That's they have to try this one out yeah that's i i think i i actually selected if

You selected gemini 3 pro in the past it will automatically switch you cool this goes to i Guess the when you're when you're purchasing meals and you're using the us you're buying our you're buying This knowledge you're buying our opinion and these decisions to make this easier for you um yeah it's Just flat out that all It's smart and it's fast it is more expensive than what i would hope which is why i Haven't replaced 3 .1 right 3 .1 is so good for its price that makes a lot of

Sense considering that they're probably using some semblance of it in order to aggregate their search results and Give you those like Gemini and answers all right so past this we go get to our flagship Models GPT for our GPT 5 .2 for it's on at 4 .6 for this like Claude Opus 4 .7 Opus 4 .8 which was launched very recently and then the most expensive model gpt 5 .5 i know uh yeah so i want to start with the important fact that gpt 4 .0

Is on there mm -hmm tell me that's an old model it's a model that you can't get On chat gpt anymore if you buy chat gpt you can't use gpt4o anymore and why do we Want to use 4o

4o is not a thinking model it's kind of it's evolved to be a bit different they kept Doing the Post training on reinforcement learning with human feedback it started feeling and getting really smart out of that Aspect it had its troubles and issues in that started getting very sycophantic you can totally drive it To be whatever the hell you want you can make it it's really good for role playing that's Where like the whole like you know gpt4o was my husband or wife and people getting married to

Them all that stuff that you probably have heard of and remember that was people using GPT -4 -0 yeah and when open AI removed it there's just was like so much loss and pain I Guess for those people and they're always like bring back 4 .0 we want something that feels human Blah blah blah this and that etc.

I included it there because I believe we're adults

And i believe that education and as long as we've got awareness that we shouldn't moralize or judge And let people use things that they make them feel good that's basically it yeah yes and we We do have this opinionation in Meos where like you're the one in control we give you the Best and we care about your experience over

Anything? Actually anything? Really actually over anything

The extents that we've gone to to try and make this good and like I guess we're gonna Do that done right series the done right series There's reasons why people haven't done this And Yeah we're just crazy enough to do it hey Yeah, it's very exciting.

What's after 4 .0? It's all of the flagships. Oh, so 5 .2, Sonic, 4 .7, 4 .8, and 5 .5. So I'm going to mention why 4 .7 and 4 .8 is there, both of them.

The reason for that is 4 .7 is a slight enhancement over 4 .6, which was a unit Step of intelligence compared to 4 .5. And 4 .7 was more directed so it will do what you say specifically 4 .8 is all Of that but they added the behavior of it is less likely to lie to you okay they In fact anthropic put a lot of effort into making sure that it is being honest and you Will see this in opus 4 .8's thinking it's like I should be honest I will do this

Thing etc it starts to say those words back to you more and more so it's very clear That they included in their post training step from 4 .7 to 4 .8 that particular like switch There and again We trust your judgment once you're educated to know what you can do and what you want to Use yeah that's a lot of opinionation and difference in results between those two yeah like there's a Element of choosing here to make it easy for for everyone to just like get going and then

Also i just don't want to infantilize people absolutely i believe that everyone has the ability to learn All this stuff It's not that bad no so we just like have a list of all the models um and We have a little description which is incredibly accurate yeah you just pick which one you want i Guess if you're confused ever you can message meos001 from the comb box and you'll like actually be Able to talk to us and we can actually give you opinions you are you will not talk

To a chat bot it is actually us we might need to put a chat bot there at Some when we like get busier but we will review all of it for sure and that but Will be trained by us and like or like yeah yeah but like if you if you download It now as like as in like when we're posting like right now

So you mentioned tool calling oh yes yes so what is that

So people may know that LLMs, they produce words and they can now I guess see that it Starts to do web searches. A web search is a tool call, it calls an external server or rather more importantly the LLM Responds back in the system and the system has to be able to see that that response is Run a function.

And then that system then calls an external server to do a web search. It comes back and spits a thing.

We should do a whole thing on MCP at some as well. Machine copy -paste. Sounds good. The tool calling that we have in Meos and on the Meos system, there are some unique tools That we've given it.

And we've put a lot of effort into being able to Make Meos this phone app. The reason why it's an operating system is that it's actually a whole application kernel and it's ai Native so the ai is able to read through your data box or have functions and tools that It can go through there and synthesize all that sort of stuff yeah so it's just like go Going ahead and finding uh the correct information for what you need that's right at the time it's Like it knows how to use your computer yeah think of it that way yeah like it knows

How to run commands it knows how to run the commands within Meos there's some really cool hyper -technical stuff about how we can extend that without sentience protocol and mcp and all that sort of Stuff where you can extend its capability of this brain but tool calling ultimately is the magic that Allows things to be relevant and contextual to you so that the results are much much better and You don't have to like copy and paste context in like memories are all handled yeah it searches

Your memories for you and i want to say importantly about Meos is that the level of privacy In this compared to anything else you're going to be using is probably unparalleled all of the data Is on your phone

They're all about cloud providers we have zero data retention policies with by default you cannot not get A provider from us we won't there's a reason why grok isn't there xai don't provide zero data Retention so everyone everything you use there they're not retaining data at all and i mean like not Even for like 30 days chat experiences do not delete your data i looked into this it's really Scary yeah the whole soft delete anyway where it's like maybe we deleted it they actually have to

Uh hold on to it yeah for legal reasons The reason why you can use these models in zero data retention is that we're effectively providing what Is enterprise -level contracts for you guys.

The same with our web search. The web search on there is using Brave Web Search Zero Data Retention. So all of these tool callings, all of these things, everything, like going over your chats, all of That is all private.

Exactly. And this is part of the reason why we have OSS 120 billion feeling so much smarter, because It gets the right context. And that's what you need, really.

Like, it just needs to know things. AI just needs to know things. It's just kind of context. You have to tell them what it needs to know.

Yeah, so zero data retention. It's enterprise level contracts. Basically, we're just saying, like, you deserve the best. Yeah and if you're really nerdy you can actually host it locally host something locally your own and Put your own endpoint in oh yeah yeah you can you can use like that as well in Like olama locally run LLM if you want yeah yeah yeah anyway we should eat hey yeah this Was fun