A strange thing about the physical ai conversation we continue to discuss smart models better robots better world models better simulation However, there's a far simpler issue beneath it all What are the legitimate sources of actual real-world data that can be used? a language model can be trained from huge text corpus due to the presence of vast amount of text The physical realm is other than the physical realm In order for a robot to function on a street, it must be able to recognize what a street is. The actual appearance of a warehouse The arrangement of objects within a room The layout of objects in a room How space is changing over time You cannot get rid of all that with scraping another site There's a part of the physical ai stack I find interesting that is involved in this, and that's it's the part I'm not familiar with. the phone is put into capture mode not a special machine in the lab merely a machine that can really perceive the world think about how much of the internet is basically information about the physical world maps photos reviews videos addresses floor plans and yet most of that information still describes the world indirectly a photo tells you what something looked like from one angle a map tells you where something is a review tells you what someone experienced but none of these necessarily gives an ai system a structured understanding of the space itself that distinction becomes important when you move from digital ai to physical ai because a robot doesn't just need to know that a building exists it needs to understand the building as a place this is where spatial data starts becoming much more interesting than another dataset on a server @vangrid_io is working around that exact layer real world capture spatial reconstruction and provenance around where that data came from

Copy and paste is great until the object you copied is where your money is going wallet addresses are too long to use by hand, most people don't check all the characters that makes for a very human problem. you copy an address you switch apps you paste it and you believe that you are on the right path to the right place Name is an argument that is passed to send.send to name is another argument passed to send. Handle to address resolution within the wallet memory layer is called the source safe architecture so rather than making users safer by having to look forty something characters each time the wallet should be able to accept a readable @name and figure out the real destination in the background That's important, not just because of the cryptography, Sometimes it's eliminating a hazardous step in the user workflow the send to name system is on in macos desktop beta and one of these little ux changes which can have a much bigger security implication @Americanfort_io

Bots have no need for sleep However, they can continue to make the same mistake @agenticscredit has a cooldown period after losing a few games where you won't be able to add any funds. It's just a blocker for execution behaviour: loss → loss → loss Temporarily put an end to trading efforts In some cases, the best thing to do is to avoid making any trades It's much more exciting to paper trade if it's using the same risk engine The paper environment in agentics credit is not a free for all backtest, it goes through documented risk controls That doesnt mean the following rules and drawdown controls don't apply paper performance is tested given constraints

there is a reason wallet addresses feel strangely easy to trustyour brain is doing most of the workwhen someone sends you a long address you arent realistically memorizing every characteryou scan it recognize the beginning recognize the ending and move onthats exactly why address verification becomes a human problema malicious address can look familiar enough at a glance while still being completely differentthis is where a readable handle changes the interactioninstead of asking someone to manually compare a long cryptographic string the wallet can work with a human readable name and resolve the destination behind the scenesthe cryptography still existsthe user just doesnt have to perform the cryptographic comparison with their eyesthats a much more interesting way to think about wallet securitysometimes better security starts by asking humans to do less @Americanfort_io

There is one simple thing that I can't seem to get my head around with physical ai. We've become very proficient in teaching machines using materials we already have online. text, pictures, videos, documents. But a warehouse floor isn't in the way waiting to be scraped Neither is an inside of a factory or as to the design of a shop or the appearance of a particular room at the present time Blind spot, that's what it is. physical ai requires data regarding the physical world, but datasets of the internet size don't naturally exist in the real world. The data layer is as important as the model layer, that's why! I'm interested in @vangrid_io because it begins with that thing that I believe is missing: Where and how do you get fresh spatial data from real places when you want to use it?

Suggestions may include a stoploss It may be a requirement for execution or it may be a requirement for the execution The second approach is implemented by @agenticscredit credit s risk engine A hard stoploss is needed when entering an order for open positions If defined condition is violated, the position will close automatically risk rules are contained within execution There is the minor aspect of how a bot processes winning trades, which is changed in this execution: If price hits the Takeprofit level the risk engine activates a trailing stop As a result, this work flow is now: take-profit reached → trailing protection activates The arrow may keep on moving as long as it is pointing towards the →If there is a reversal , there is an exit If there's a reversal, there's an exit Its a rule and not a manual decision

you know is making address safety way easier to understand with @Americanfort_io Its comforting to visit the initial and final few characters of the address Matching 0x1234...5678 doesnt mean that you have matched the entire address That's where the ability to spoof an address gets interesting Send to name is an alternative to the “eye” game that has always been part of sending to a name; it makes sure that there is a verified flow of @name ne handle multiple networks Send to name only supports wallets for btc, eth, erc20 usdt, usdc, 0g and ltc in the macos beta It's really an easy concept: Don't require users to have to deal a separate string for each payment

Let me tell you Consider what it is you are seeing versus what you are understanding. Given an image , a vision model can recognize a chair. However, a robot would require additional information in order to interact with that chair Are you 2 metres away? Does it block any way? is another object behind it? Does the robot have the ability to navigate around it? what is the relationship between the chair, the floor , the walls and the rest of the environment? This is where spatial intelligence begins to get out of the ordinary image recognition the physical world is relational Objects are not independent a chair is in a specific location within a room The room is located within a structure the existence of the building at a specific site and all of those relationships may be significant to a machine that has to navigate or interact with the environment This is why data is more complex with physical ai than just taking more photos a picture can show a model what something looks like spatial data can give an impression of the location of things and the structure of the environment @vangrid_io is creating on top of this spatial layer of data, starting with the real world captured data the larger concept is that a machine would require representations of places, rather than representations of objects. The data problem is shifted from “what am i looking at?” to “where am i, what surrounds me, and how are these things related?” with physical ai.

Paper trading can seem to be a great experience. But the reality is quite another when you get to then real execution. @agenticscredit gives greater weight to real onchain When it comes to trading history versus paper performance, it becomes confusing. that makes an important distinction: Context can be developed through simulated performance. real execution is more weighty The scoring engine needs: 20+ closed trades 14+ active trading days Prior to giving an official rating That's a simple concept with a significant implication: Creditworthiness requires some track record of worthiness

The fun part about send to name: The name doesn't need to be a single public receiving address Instead, a destination could be generated by the wallet for the particular transaction Same human readable handle Different receiving address for the transfer @Americanfort_io