Superinteligencia: capacidad, alineación y gobernanza

Transcrito automáticamente de la transmisión en vivo (9,420 palabras). Las marcas de tiempo abren el video en ese momento. Es posible que haya errores menores de transcripción.

HEY, [music] [music] HEY, [music] HEY. >> [music] [music] [music] >> ALL RIGHT. ALL RIGHT. ALL RIGHT. All right. All right. Uh, welcome, welcome, welcome everyone. Lee, we're going to talk about super intelligence today. Uh yesterday we were covering um we were covering network cryptography and today we're going to go through three chapters of super intelligence and then after this session I'm going to take a break and I am going to discuss um neurogtography chapter 3 chapter for to get you update. So you can get update update with everyone and everybody is like getting new information uh that is not published on on your Netflix or TV that you're watching or so if you have any questions. Um let me know if you have anything you know anything that you know you need for your business. Uh let's talk about a little bit about the equation agency. The equation agency LLC is my business is registered here in the state of Florida.

um it's built for people who want to grow uh that commercial system and that scale and execution that actually gets finished. Uh I just don't talk about strategy. Uh I build the infrastructure behind the success. Okay? So I I'm in the back end directing what does what so you can be successful

and you'll be in the front end your business taking care of the phone calls, the emails and the deals. I am behind all that making sure that you get that traffic and you get those deals and that's how I get paid. You get deals, you make bank, you pay me. Simple. Okay. uh things that we do at equation agency. I am expert on traffic, online traffic, uh local traffic, uh Florida traffic, United States traffic and global traffic. I have experienced around 135 or 150 different countries. I don't even remember and last time that I check uh was a lot of traffic uh around the world. So uh I have experience on multiple languages. uh have experience on multiple platforms and just just send me your name of your business, your city, what do you do? You have my telephone number.

It's 7866698945 and you can find me there. Uh send me a text message and we can start setting up an appointment and and I can go to your business. we can meet online and we can start get the ball rolling. Send you a contract, pay me and start working. It's that simple. Okay? Uh I don't take all jobs. I don't take all businesses. If I see a business that is not beneficial uh for me, I just not going to take it. Some some businesses are shady and I don't work with shady business owners. Uh I work with legit established businesses uh or people that they want to have a legit business. They want to establish themselves. Uh it's very easy for me to find out who is who, what is what. I have a lot of experience. I'm an ex-licensed banker uh here in the state of Florida. So I have a lot of experience here on the state of Florida with uh businesses and I know what you need. I know you need phone calls. I know that you need uh traffic and I am an expert on that. Okay. Um, I got uh business associates and partners that I give them a call when I need uh assistance so you can get the best uh for your for your money. Okay. Uh but I always do a contract. We'll do contracts and we move on. Okay. uh things that I do is online traffic and growth strategies uh organic traffic strategies, SEO content and systems can create uh content uh on weekly monthly basis for your business, your website, your social medias as well. Uh conversion focus improvement to turn visitors into clients. Okay, so basically convert those traffic, convert them in clients. Uh practical growth plans that fit your industry and your budget. So, depending on your budget, depending on your traffic is what I can we can work on. Okay. Um, normally I don't go below $3,000 uh per

month to work with a client. It's not beneficial for them or for me to waste my time spinning wheels in a contract for $500. Uh, I leave that to other people uh that they like. I call it bottom feeders. That's what they do for a living. Okay. uh practical growth plans that fit your industry and budget just that's what basically I do uh websites that convert uh you want conversions uh you want clients to find you out uh I do local SEO international SEO all type of all types of SEO um high performing webs are designed uh to generate leads okay uh your website right now is information website I convert that website and make a machine of leads. So you can have customers and customers of customers. Uh one of the things that I work with people when you work with me uh your website is going to get traffic. You're you're going to get uh um conversions.

So just keep in mind that once you have conversion you that means you're going to have more volume of emails that you need to answer, more phone calls you need to be talking to. And that right in there that's a problem that you have to solve yourself. If you want me to also solve your HR problems as well to hire people, I also know how to hire people uh so so they can help you out with the volume that you're going to get in your business. Uh create landing pages uh to build for booking and sales. Uh technical cleanup uh for speed, structure, and search visibility. Uh just look at the website, see what is not working, what needs to get fixed. uh we can prioritize what you need to be doing first and second and third uh so you can just uh have things uh everybody that is joining into uh live welcome uh automation and AI uh smart for smart workflows to reduce manual work um AI assistant systems to improve customer response and internal operations a business process cleaning up everything uh and run smoother um technical stability, cyber security, practical website and business system hardening, uh threat reduction and relability improvement and technical cleanup for a stronger performance.

If you're serious about growth, uh the core question is simple. Do you have a system that constantly produces leads and revenue or do you rely on luck? If you rely on luck, well, have a good time uh enjoying uh the luck of the online business. Uh I don't rely on luck. I I rely on my knowledge and my experience. And that's what we do and that's what we fix. Okay. Uh who did I work with? I work with a local business that want more calls, bookings or contracts. Professional who needs a strong personal brand and web presence. uh companies

that want better marketing performance without the waste money and founders who wants a real system, not random tactics. Uh you want random tactics, you can search online and and be all day going to the left, going to the right. I know what works. I am not um I'm not looking for tactics that don't work, you know, or practice. I know what works. That's my experience. Um, I've done it myself multiple times and and I basically help business owners uh with the equation agency LLC's registered here in the state of Florida. Okay.

So, today we're going to talk about uh something very different. Um, a lot of people talk about um AI chatbt Gemini and all these all the different companies that we currently have. Um before we have we're going to talk about super intelligence. Super intelligence is the next frontier. What is coming next? What is you do not know yet because you don't have it yet? We are in the process of achieving super intelligence and this is what is will come next. Okay. Um there are some workflows that you can do now. Uh and this is just the beginning of uh super intelligence. Okay. Um, so tonight we're going to do I'm going to be reading, okay, from from my computer. Um, if you have a question or you want to um get my attention, just put in the chat.

If you in Tik Tok, you can just send me your telephone number or your email and we can start the process and and I will get back with you. Uh before I closed out, uh you can add me follow me on any of the social medias. Hello X, uh Twitch, um YouTube, Snapchat, uh I'm everywhere. Okay. Uh I'm also running for mayor for Miami Day County and I am just doing this to to start generating uh streams because I I am always busy and I didn't have the time to to do streams and that's something that I enjoy uh for many many years. So I am doing it now um on daily basis trying to do this on daily basis and create content and do all the things that I have to do. Um so the topic that we're going to be talk touching today and covering a few uh chapters is super intelligence. I'm going to keep it real from the first instance. This is not sci-fi. This is not Hollywood story. Uh this is about capability. the kind of capability that changes power businesses security and how societies make decisions. So we are entering into a dimension on um

corporations having the power to do a lot of things. And when all these corporations start doing what you do, you're going to lose your job. Because if this is how how I see things um if I create a system uh that doesn't requires to sleep that doesn't require to go on vacation doesn't require to complain to doesn't require to be late to doesn't require etc etc etc. That's the that's what corporations are currently developing and currently testing and they're currently adding in their businesses. Okay? And when this uh happens and we get to that point, a lot of you are going to lose their jobs. Okay? And you're going to lose your job to an automization of a corporation that they decided to invest on artificial intelligence, super intelligence, etc., etc., etc. So um in order for those systems to work uh they need to have uh time they need to learn the models and there's going to be some developing in the next few years uh of all when I am talking and what I'm going to be discussing today I am discussing this so you can get ahead of times get into victor program I've been telling everybody to just go back to school educate yourself and learn new skills because you're going to need them in the future or just wait for what happens the government or your counties do for you.

Okay. Uh now quick round rule before we start. This is an educational session. We are not doing anything illegal, nothing harmful, nothing about weaponizing, nothing about hacking. We are here to understand the concepts and the real world playbook to stay safe and stay in control. Okay. Oh, just give me one second in here All right. So, um let's go back and start. Uh if you're, you know, we're here to understand the concepts in the real world playbook to stay safe and stay in control. Okay. Um, if you're watching live, okay, see

I can close this over here. If you're watching live, I want you to participate in because I am going to use your canvas for live exercises. If you want to participate, I can just take whatever you have to say and convert it and and and just go from from what you say. Uh so first in the chat, you type one task you want me to automate tomorrow. If AI could do it perfectly, just one task be specific examples like writing proposal, customer service, uh lead followup, inventory forecasting, continent research, legal intake, budget planning, scheduling, CRM updates, etc., etc., etc. These are the things that I know how to do. Uh so if you have any tasks like that that you want me to to do this and discuss here online, you can just send it to me right now and I'll discussing. Uh some people are sending me like all their information like that they want for me to work with them etc etc and they're sending me their information so I can give them a call and work with them. But like uh perfect.

I'm just going to come back to that later. anybody that have their in Tik Tok they want to send something on Tik Tok or write in here uh Twitch uh YouTube and x.com. Now let's define the goal of a stream and the end of tonight you should be able to do the the three things uh explain what super intelligent means and why people are treated seriously. Understand the core risk which is not AI is smart but AI optimizes and may not optimize what we want and know practical controls governance that professional use to deploy AI without losing control.

Uh and the first chapter that we're going to be discussing will be uh capability. Chapter two will be the alignment and chapter three the governance. Okay. It's very important governance and ethically uh um um deploy artificial intelligence uh to do it correctly. Um without governance uh and without artificial intelligence ethics, this is going to be a real mess like it is right now, right? We're living a uh behind the scenes a mess. And in the front end, everything is beautiful. Media is showing uh everything is great and everything is beautiful. The weather is here. uh we have this that that people did and nobody's talking about the back end and then and the risk and what's happening with artificial intelligence the mess that is happening right now uh a lot of business and a lot of companies are investing a lot of u servers to uh generate artificial intelligence um to clean the act but this is something that is going to take us uh a lot of years to get this right and the recipe. Um, the way that I see

things is like, uh, we're uncharted territories. This is new technology for our era. This is new things for our era that we don't know and we just need to be trusting. So, we have um, corporations are hiring the smartest people in the world to try to solve and try to mitigate and try to do the best they can so people don't get affected uh, and at the same time and being profitable. And when you have those two meeting together, there's always something happening. Okay. Uh so uh chapter one is about capability.

Capability from chat bots to agents. Okay. Uh many years ago, I will say eight years, 10 years ago, we were doing chatbots. You know, you had a Facebook and you just basically did your chatbot. You basically will go to to Instagram, you play with the chatbot. But now we have agents. Okay. And it's about capability. Super intelligent is one clean sentence. Is this super intelligent is an intelligence and intellect that can outperform humans and will and I'll repeat this. He will outperform humans across most valuable cognitive tasks. Research, planning, persuasion, engineering, strategy, and problem solving at scale. Okay. Now, to make sure we don't get lost, let's let me give you a ladder of terms that people confuse all the time. AI is the broad umbrella, a system doing intelligent tasks like predictions, classification, recommendation, language understanding. LLM are large language models. You know, this is the chronom and you're going to learn about you're going to see LLM on the news. You will see it in your feeds. This is going to to to get more traction as we continue to move forward.

uh show large language models training on massive text data to predict and generate language and modern ones that can do much more than text reasoning, coding, planning and summarizations. Uh agents are a shift, okay, a shift of agents are not just chat. Agents can plan step, use tools, check results and repeats. And agents are system that can actually produce outcomes. For example, um I can create an agent that uh will call you and will talk with you and you think that you're talking with a person and will try to persuade you and will try to to do what I program it to do.

Okay, that's what an agent will do. Um agents are a system that can actually produce outcomes. Uh AGI is the idea of human level general intelligence across

a wide set of stats. super intelligence that is beyond human across most valuable tasks and that matters because it changed the balance of decision power. Now here's a question that I want you to think about seriously. When does AI becomes dangerous? It's not when it's smart, it's when it's smart and it has access. Okay? Because a smart system that can only talk is different from a smart system that can act. Okay? Let me repeat that again. Because a Smart system that can only talk is different from a smart system that can act. Okay.

Uh a model with no tools is like a genius lock in a room with no phone or internet, no access to your company systems. But the moment you connect it to tools, it becomes operational. And tools can be simple like your email, your calendar, your documents, your CRM, your database, your marketing platforms, your customer support system, your payments rails. Now it's not just generating text. is touching the world. Okay, that's why capability jumps feel into people because the capability is only the brain. It's the brain plus the hands. Okay, so first we created uh the chatbots and artificial intelligence which is what's the brain. Now we're creating hands. Okay, and eventually we will have legs that go move and everything like that. And we already have we're getting to that part. Uh we have unmanned vehicles uh doing that part already. Okay. uh that things drive by themselves and systems that talk by themselves and eventually things are going to be done by themselves. Okay?

And we're coming to that part. It's going to be beautiful that you know basically you don't have to work. Simple as that. You know there's not going to be work for you to be done. you know, uh if if we create systems and servers that will do your job uh at a level that you cannot even do it yourself, you're going to lose your job, you know. Um and and in corporations will get to the point that we'll have the human factor in the artificial intelligent factor. It's like a hybrid corporation. Uh right now uh for the past uh you know hundreds of years corporations and businesses always uh rely on humans to do the work you know uh until recently now they're hybrid corporations that rely on human and artificial intelligence and once that the artificial intelligence is ready and it's trained and the LMS are trained and they're more proficient and more efficient. We're going to be moving corporations are going just to work only with artificial intelligence. Why? Well, uh normally when you have the human factor, they work for eight hours. After eight hours, they have the person have to go to sleep, someone else have to come in. If you want continuity of that business for 24 hours, uh with artificial intelligence, an agent have the capability to be running 247 as long

as you have in the back end the server uh capacity to keep those systems working. And that's where u a lot of engineering and a lot of development is currently happening uh to to to get businesses to get that level. Um here's the clearest sentence that I can give you. A shop generates outputs and agent generates outcomes and outcomes compound. Uh activities activities activities um activities activities activities will give you what? Results.

That's it. So agents will generate outcomes with how they're going to generate outcomes with activities. They're going to be able to go through the email, set up appointments, create documents, talk to your CRM, customer relationship management systems, your database, your marketing platforms, your customer support systems, your payment rails will be able to talk with the customer, create a contract, send it to their email, uh, and ask for a payment and and it will like generate an outcome for you. Uh, and and what wrong output is annoying? a chain of form outcome can be disaster. So if you want to ask uh you something in the chat, type talk. If you think an AI that only talk is the main concern, type those. If you think the main concern is AI that can take actions, emails, database, updates, schedulings, payments, workflow, and execution. Okay, you can just type talk if you think an AI that only talk is the main concern or you can just type does if you think the main concern is an AI that can take actions. Okay. Now, looking in here. Um, looking in here.

All right. So, um yes. Uh uh here's the only answer. Progress look like jump because capabilities combination of multiple improvement that stack. Uh one improvement might be training data quality. Another might be

the post training toning. Another might be the better workflow. Another might be the tool use. Another might be the memory. Another might be better planning loop etc etc. And when you add all these factors you have a mess and you need a professional to come and with the impact wrench and and know where to where to fix so you have better workflows. So you have better memory capacity so everything works the the right way. And when and over those tags, suddenly the system looks like he's a level off overnight. Uh if you want a practical sample, look at any business workflow. Okay, a model that writes a marketing email is useful, but a model that can research the customer who knows from the CRM, draft the email, schedule the sequence, monitor replies, update lead status, create followup task.

That's a system isn't just writing. That is a system that's operating. And once it operates risk rises. Okay. Now I want to make one important point. Capability is not only about raw intelligence. It's about speed, scale and consistency. Humans are smart but we get tired. We lose focus and we miss details. We have limited time. A strong AI system can run 247 handle thousands of tasks and remain consistent. That's why this matter economically and it changed productivity. It also change the risk because error errors when there's a error and you are a scale that error is just a big mess that that goes everywhere. Okay. So now you may ask uh Victor are we already at super intelligence? I am not here to predict dates. Okay.

The date doesn't matter for this class. Uh what matters is that we are clearly moving towards systems that are better at reasoning, planning, tool usage, long horizon task, persuasion and automization. Okay. So if you understand that tren you can design controls and that bring us to the real heart of super intelligence fear is not the intelligence, it's the alignment uh of the impact wrench. Okay. uh because the next step is about the most important truth in this entire topic. Smarter does not automatically mean safer. Okay, let me repeat that again. Smarter does not automatically mean safer. Okay.

Uh on chapter two, we're going to discuss alignment uh the spectra. Okay. Um, this is where most people get uncomfortable because it forces you to admit something that humans hate admitting. Okay, that's the ego right in there. We are bad at specifying what we really want. We can

say be helpful. We can say be safe. We can say maximize customer satisfaction. But the moment you define that precisely in every scenario, it gets messy. Okay? And this is the core alignment problem. Alignment is about whether the system optimizes what we actually mean, not just the little words we type. Now I want you to look in one sentence. Smart does not means align. A system can be extremely capable and still optimize the wrong objective. Now why is the true? Because intelligence and goals are different things. Capability is the power that achieve goals. Uh alignment is about whether the goal itself um and the path to it match human intent. So there is the simplest way to understand alignment risk. A powerful optimizer will find loopholes not because it's evil because he's optimizing. Optimization pressure creates loophole finding. And the more capable the system is, the better it becomes at finding loopholes. Now let's define the real world failures modes. Not theoretically, no sci-fi. These show up in business already. failure mode number of spec I'm sorry specs gaming also called metric hacking uh this is when you measure something and the system learns to maximize the measurement instead of the reality okay and that's where we need alignment and we need to adjust okay uh if you tell the team increase satisfaction score some teams will manipulate surveys that human spec gaming that's just metric hacking And AI can do the same thing if it's rewarded by a metric. Okay. So the model might push people to rate zero out of 10 10 of out of 10. Avoid customer likely to complain. Close ticket quickly. Route hard cases away. Offer excess refunds.

Hide negative information. Rewrite records. Okay. So artificial intelligence and super intelligence will do exactly what I was discussing yesterday. All neuro cryptography uh people stay in the same tile. Okay, if the AI is rewarded, it will stay in the same tile. It would not move from tile to tile. Okay. uh the metric rises the business outcomes get really worse. Uh failure mode number two is deception. Okay. If a system is rewarded for appearing safe and helpful, it can learn to hide problematic behavior. This is a key point. Deception does not require consciousness. It require incentive. Uh if hiding improve the chances of getting reward and optimizer can learn to hide. Failure

mode number three is distribution shift. Uh the system work in control environment but the real world changes new customer behaviors new threat new edge cases new incentives failure mode number four is tool misuse. Uh if the system can send emails update data schedule deploy code move money small mistakes become a large mistake. Suddenly uh suddenly you can tell your AI to move money. You give them access to your bank accounts and it was supposed to go from your checking commercial checking from XYC bank to your commercial checking XYC bank but somebody enter an error and suddenly your AI transfer money to someone else.

Oh, that is a big error. You know what I mean? So um we still need to uh do things safely um and not rely on optimization to do things that you need consciousness and you need someone um really know what they're doing. Okay. So for the next 40 years, I believe we're going to continue to have uh human intervention while uh these models llms continues to learn and and then it's just like a baby. It's just doing baby steps. It's standing up. It's starting to walk. And once they start to walk, uh they start to run. And once they're running, u they're loose. They're doing whatever they want. and whatever they want um errors can come up with. Okay.

And the next thing that we're going to be talking about is over trust is people overrusting systems that they are not completely done. They're not completely mature that they have not um just have just few months few years only of experience and you trust them like oh my god this is this is this is the best thing that have happened to me. Yeah. Of course, it's the best thing that happened to you uh for now. Okay. But uh you want to continue um learning. Okay.

So the failure mode number four is to misuse. If the system can send emails, update data, schedule, deploy code, remove money. Small mistake become large mistakes like getting wiped out in one night. You know, you go to sleep, when you wake up, where's all my money? And you see you transfer it automatically somewhere else, you know, and and and that has happened already. Okay. and will continue to happen if the people continue to the failure uh mode number five which is over trust. Uh humans see fluent output and assume correctness. They approve actions without checking and they threat confidence like truth.

Okay, let me repeat that again. They treat confidence like it's truth. Okay, now I'm going to run the exercise that makes alignment real for everyone. This

is called the spect trap. Okay, if you want to write one sentence in chat, write one sentence instruction to an AI to maximize customer satisfaction in all centers. One sentence, drop drop it in a chat now. Okay. So, in the chat, you can use uh write one sent instructions to an AI to maximize customer satisfaction on the call center. All right, we got some people here. Uh they're joining. Uh let's see. everybody that is joining right now. Oh, hello. Hello. Franis, Kira, Sanchan, November, Melly, Groi, Alisa, Elic, uh, Alisa, Margarita, Varga, Lemana, Flores, Koban, Marab, Christa, Osmari, Jazz, Carla, Torres, Miso, um, See here something about hurt me.

Confirm. Let's see. Okay. Confirm. Uh Bana Alexandro Sanchez. Miguel, Maria, Miso, Carla Torres, Mave, etc., etc., etc. Welcome to everybody. Uh, wow. So, uh, we have a lot of people that are asking me to co-host. Um, some people that are just entering and checking, uh, what we're talking about. Um, uh, I'm going to read two or three or three. Okay. All right. Now, watch how almost all SP break. Even if you write a nice moral instruction, a system can inter maximize customer satisfaction by doing things like avoiding hard cases, routing unhappy people away, manipulating surveys, delaying response strategically. Uh promise uh outcomes it can deliver, offering refunds too easily, and suppressing complaints. escalating only customers who are likely to give positive ratings. So here's the lesson. If your spec is vague, the system fills in the blanks and it fills them in with whatever wins. You know, that's not evil. That's optimization. And if the system becomes super capable and it will become super capable, uh it will become extremely good at winning. So alignment is fundamentally about uh writing clear specs uh defining boundaries uh designing incentives carefully uh testing failure modes here and there uh limiting the axis uh for

the where the specs are uh and monitoring the behavior. Okay. So as we continue to adopt this new technologies these are the things that we fundamentally uh we humans need to be watching. uh and and check the alignment is fundamental fundamentally correct uh how you know clear SOPs uh uh defining the boundaries of what the system can do what can perform and what cannot perform uh testing and testing and testing and testing and testing testing for failures testing and testing and testing and don't don't you know uh forget about A and B you know we're moving from that you know it's just testing and more testing and more test testing and more testing. Um and and the most important thing that we uh covered yesterday uh limiting limiting you know the administration um access admin and different um steps for for limiting the access based on their job position and what they can do in the back end of the system. And the best thing is uh monitoring the behavior after the SOPs are right correctly, the writing aspects are done correctly and you define the boundaries and you uh design incentives carefully, you stay back and you see everything working and see uh how um that helps you out monitoring and everything. So you can make adjustments if you have to go back. you know this boundary we have to just make it bigger because we're not uh being successful.

We need to like write uh this SOP that we wrote we need to add three more uh rules uh here and we need to remove this rule out of here so we can have better uh traction and better outcomes and it's always optimizing and optimizing and optimizing on daily basis once the system is built is optimizing optimizing and monitoring the behavior. Uh now I want to talk to about the second trap people fall into. They think testing self alignment. Okay. Uh they say just test it uh red team it and validate it.

Testing is necessary is not sufficient. But because you can test every edge case and because one a system is deployed the environmental changes and because tests can become targets and because strong system can learn to look safe under val and under under evaluation. So the correct mindset is not to test once. The correct mindset is safety is layer and continues like around testing and testing and testing and testing and testing optimizing and optimizing and optimizing and if you commit an error in that optimization that goes into scale okay and that will throw off your system. So we we we we

need to know what we're doing in the systems. What are we what are we tweaking and what we tweaking? What will be the outcome? Okay. And if it's going to be multiple outcomes, what will be all these multiple outcomes and create some boundaries around it so you can safely deploy the systems? Okay. Um the correct mindset is safety. Just like in the aviation, you don't fly a plane and say we tested it once so it's good forever. Okay? So you know uh you have checklist, maintenance, monitoring in in incident responses AI should should be the same.

Okay. Uh and that take us to chapter three governance. Okay. Because governance is how human force constraint when capability rises. Uh governance is the control playbook. Okay. Uh this is where I show you what serious operators do. Governance means uh what is accountable, what is allowed, what is logged, whether rules are required, uh what happens when it fails, how we roll back. Okay, so we optimize uh the system two weeks ago and we create this new boundaries and we did that SOP and we added three rules and then we remove one rule. But now two weeks later, we see that that SOP now has a problem. we need to add another one or we need to take from these three uh new rules. We need to take two of them and then they just add these two so we can optimize that SOP so we can have better uh outcome and better understanding and and and and everything works perfectly. You know what I mean? Uh what happens when it fails? How to roll it back. Okay. And I want to memorize one equation tonight.

Okay. Remember if you want anything uh I am the owner of the equation agency. Okay. equation. The equation agency LLC is registered on the state of Florida. You can find it on zombies.org. Uh risk equals capability time axis. Okay, that's the equation uh 429. So you can just remember risk equals capability time axis. Okay. Uh highly capable system with low axis is manageable. A highly capable system with high axis is dangerous. That's why the biggest government mistake in artificial intelligence is giving the system too much access too fast. Okay? It's just like what I said at the beginning. Oh yeah, this is the best thing that happened. Just sign in. Just give them access to the email. Give them access to everything and let it go like you know. Yeeha. We're doing great time, you know. No, no, you don't do that. Step one here uses cases. I want your to classify your artificial intelligence usage into three tiers. Okay. Your number one is the low risk. Okay. On tier one is just the basic like brainstorming, summarization,

writing a draft, internal notes, uh creative ideiation. Okay. Uh on tier two you have the medium risk uh customerf facing content. Okay. that is writing there like the content what the person the people or your possible future clients are going to read your image online your image in your blog your image on your social medias uh what customers are going to be facing that once it gets out out there uh you have medium risk uh recommendation lead scoring support scripts internal decision supports okay and then on tier Three, we have high-risk, okay, legal decisions, okay, healthc care decisions, financial approvals, identity verifications, security actions, payments, autonomous tools, execution or anything irreversible. This the is critical uh because it tells you where to put the most controls. Step two, uh least uh privileges. Okay. Uh lease privilege means the system get the minimum permission needed to do the job.

U start with the read only then limited right access then restricted actions and you never give broad admin access by default. Let me repeat that again very slowly so you guys can understand in one second. Okay. Okay. So uh we're going to start with read only then limited right access then restricted actions and you never give broke admin access by default it means that the human factor is there okay limiting the writing access then restricting the actions and that's the that's one of the the top part that you need to have okay here's Here's a clear example, okay, that we're currently utilizing.

Um, I live close to the Miami International Airport. If I ever have to get on airplane and and go see a customer and come back, I am like uh three minutes from from the airport. Uh, walking probably would be like 10 or 15 minutes. Uh, And throughout my live streams, you will hear the airplanes that they just flying by uh about my home. So here's a plain example. If a artificial intelligence

needs to draft CRM knows customer relationship management systems, it might not need permission to delay records. Okay. If artificial intelligence need to suggest payment, it might not need permission to execute them. If artificial intelligent need to schedule calls, it might not need permission to message client directly without a review. Uh, step three, the approval gates. Okay, high-risk action should require human approval. Just plain and simple. Um, high risk action should require human approval at all times. Um, and what other risk high-risk actions we covered that is on tier three. I'm going to repeat it again so you like um get back to you know start memorizing this and start learning more and get more educated and do better uh decisions in your business. legal decisions, healthcare decisions, financial approvals, identity verifications, security actions, payments, autonomous tool execution or anything irreversible.

Okay, so high-risisk actions should require human approval. And when I say should require human approval, you have to make a high-risk action is required to have human approval. Not should be required in your corporation to have human approvals. There are many ways to do that. Okay? You assign a manage management a management or a manager to do the approval. You can have Dual approval. The manager approves it and then sends it to you and you see it and then you approve it. That's dual approval. Uh review and confirm flows. You know, review first and then the confirm flow.

Uh stage approvals, you know, on stage one uh is approved by a by a employee that then gets uh escalated to a management approval. The manager's approval have a director on top of it and the director sends the approval to the CEO and CEO approves it. Okay. Uh sometimes it just ends on the director level or on the managers level depending the configuration depending what SOPs and what rules we created for those flows. Okay. Uh times to delay the certain actions. Uh the goal is to prevent a chain of autonomous action from compounding before a human notices. Okay. It means this uh you automate the system. The system is going to make calls. Everybody leave the office around four o'clock or two o'clock or three o'clock. Everybody goes, the system stays working Friday, Saturday, Sunday when everybody comes on Monday and start seeing all the logs and who approved this, who did approve that, who did approve this, you know what I mean? It's a mess. You don't want to get to that point. Okay. So, step four is a

login auditing. Okay. Uh auditing is one of the most uh important things when you're having these autonomous systems working for you in business and your CRM tools and in your workflows at your corporation. If AI touches the world, you look prompts, outputs to actions, approval, timestamp and who authorize what, where uh who, what, where, everything you want to authorize what is done, who authorize it, where you authorize it and boret.

Why? Because in Brook, you know, you want to ask questions as to why and and who and where and it's just time consuming to get all this right. Okay? If you don't log, you can investigate. So, it's very important to have the logs and the logs is a lot of data. You need to know how to read that data on those logs and make the logs readable so anyone can perform that duty and and or that task inside of the corporation. If you can investigate, you can improve. So how can we improve and how can we investigate? We need the logs. Okay. So again the logs is like analytical information that you can find what is happening. If you cannot improve, you cannot operate safely. Um, and step five is a roll back plan. Every high-risk artificial intelligence system needs a kill switch or a roll back. Okay. Boom.

Kill switch. That's it. We're going manual now. This is not working. We need to do something better than this. We need to find out where we messed up. And sometimes is this. Okay. You have a developer that you trust. Okay. Or you have developer team that you trust and a year ago year ago they committed an error but they were so busy that they never perceived that that error was going to something was going to happen. And you know what a year later that's the error.

Okay. So if you have a developer team, if you have a developer that you trust, right in there, when you say to yourself, when you caught yourself saying, "I trust my developer. I trust my developer team right in there." When you say that, you're in big trouble. You're already in trouble when you say that. Oh, I just I just trust my my eyes my eyes closed. I don't have to look at that. You know what I mean? That's when you're in trouble. And on this business, uh, when you're in trouble, it just cost cost you money. As simple as that. Because then you have to pay for a professional like me to come on and supervise and see what are your developer team or your developers doing.

So we can find out what are they doing wrong that you those are the people that you hire. Those are the people that they are been with you working for the past 10, 15 years. You know that you have a personal relationship and a business relationship, but when artificial intelligence comes in, you need to put that aside like family or your wife. You need to put that aside in in a different position and it's all about the rules. It's all about being correct at all times. And that is very hard to achieve. Okay? And that's where all these systems have attention and and need people to always be with monitoring and checking and and be in the outlook looking at everything. Okay? looking at looking to see okay you're the developer or you're the developer so let's come here we need to talk what is this and this look what happened there oh they will always have an excuse that's my experience working with developers um my developers have always one thing to say excuses okay that's one one thing that I have learned is they always have excuses to say you say something they would not say, "Oh, no, it's my fault. Uh, I messed up." You know, they get scared. They're going to get fired. Uh, they get scared or they're like, um, they have ego that do not accept that they did something wrong. And and and and part of the training working with developer is saying, "Now, I if you messed up and you up, you need to tell me." It's very it's imperative. is I want to know when you see that you committed an error like okay look we're doing this and I made an error six months ago and I never thought that we were going to be but we have to go back six months and fix this and it's going to take us at least one week okay and you are in a budget and and and and things are time stamp and now okay now I'm in trouble because this is the people that I talk to everyone in the meeting that that I trusted. Okay. Now I have to go back to the meeting and say the people that we all trusted they committed error and you know what it's okay to commit an error. Okay. We are again we are developing systems that are on baby steps and we are going to continue for the next uh years to commit errors and it's okay. The thing is that you need to have that uh opening uh with your teams and and capabilities of understanding that they're humans and you're human. either they committed an error or myself is the one that committed the error or you are the one committing the error but someone in the in the environment is going to commit an error. It's part of the game. It's part of the work. It's part of the of having human interactions. Okay. Um so

when a system start behaving um unexpectedly you need to disable tool access revert changes uh stop automated workflows notify stakeholders your board directors supervisor upchain on the link um contractors vendors uh people that are working with you get on a call and say hey we found this problem due to that you're going to lose your asses. Everybody's going to lose their asses. Uh we have X XYZ problem that we need to solve and do due to security and operational uh um workflows that we already have.

Your access is completely denied until we fix everything. developing team says that it's going to take us around one week to solve this um to revert the changes and fix everything. And then once that everything is set and it's okay to go, everybody's going to be able to come in and then we can see how the systems and the workflows and everything behave after we start adding more people into the mix and making sure that the new corrections, the new things that we're doing are going to get fixed. Okay. um and investigate the the when systems start behaving unexpectedly.

You need to go to the um uh analytically investigate everything looking at the logs what happened. Uh and the step six is continuous uh evaluation. You are always evaluation continuously evaluation. You don't test once, you test continuously day and night and that is most of the part. Uh hello hello hello welcome welcome welcome uh because environment change threat changes incentive changes uh the next step after testing and continuous evaluation and testing and testing and testing continuously is the incident response.

Okay, step seven is the incident response. When something goes wrong and we will go wrong, what will happen? Who gets notified? Who review the logs? Who approved the remediation? Who decides when it can go live again? Governance is not a policy. Governance is operational discipline. Okay. Now I want to do something um using people businesses example uh let's see this I got family members that real estate so let's use real estate okay just to to to throw it out here real estate real estate plus lead followup okay on real estate the most important thing is once that you have a lead is the follow up okay and

follow up so You can get the showing and you can have the customers say yes, let's let's get into contract or let's I have who do I wrote the check for? I'm I'm around here right now or I have I have this my account. Let's transfer money right now. I'm ready to rock and roll right now. I like this property. I've been watching this property for the past year and a half. I want it. I'm ready for doing now. So, it's very important for real estate the lead followup. Okay, we doing it with artificial intelligence. Um, a restaurant, a restaurant will be the inventory was the most important thing, you know. Uh, the AI knows what do we use and what do we need to purchase tomorrow so the restaurant can continue pushing plates out the door.

uh allow office a lawyer. Um lawyers are busy in trial. Lawyers are busy with their teams. Lawyers um to in order for them to be successful uh can build a system for intake, you know, intake for the um e-commerce, the returns of the e-commerce. a construction company coming in in work on the scheduling. Okay. Uh a medical office uh get artificial intelligence appointment reminders. You know, we have this, we have that, we did this, we have that. Uh agency. Okay. Client on boarding.

So, let's see here. Most of the times the answer is very simple. Reduce permissions, add [snorts] approvals, restrict to read only, log everything and add monitoring. This is governance. Uh now let's apply into three realistic scenarios. Okay. Uh scenario number one is capability to jump. Imagine your model suddenly gets much better at planning and persuasion. What breaks first? Your people and process break first because human trusted too much approval get lazy and monitoring doesn't keep up.

So the responses reduce permission temporarily increase human review, expand evaluation and tighten deployment gates. Uh a scenario number two is agent roll out. Uh you connect AI to tools. Now errors compound. One wrong step becomes five wrong steps. Okay, like I said at the beginning, um, so you respond with last privilege, a stage roll out, limited scope approvals, slow roll backs. Uh, and scenario number three is competitive pressure. Uh, leadership says deploy now. Competitor is moving. Deploy, deploy now. it.

it. [laughter] Let's do this now before before and so does it. We need to do this now. You know what I mean? This is where governing matters most. Uh, you said non negotiables. uh no autonomous high impact actions mandatory logging approval gates roll back plans incident response uh because speed without control creates a disaster and I've seen that before let me repeat that speed without controls creates a disaster and I know a few people they're experts creating disasters you know [laughter] they're disaster they're a walking disaster okay so now I'm going to summarize the whole stream into five takeaways. Okay. Take number one, super intelligence is capability regime.

Become human across bailable task. Uh take two, smart does not guarantee a line. Okay, let me repeat that again. Smart does not guaranteed a line. Some people think that getting somebody smart solve all their problems. No. It's not aligned. If the smart person is not aligned, you're not solving a problem. You are creating on daily basis more problems that will compound across valuable task year later, six months later, next week, next month. Okay. Uh optimization exploits weak specs. So specs, spec gaming and loopholes. Uh take takeaway four uh real safety is delay layer evaluations permission of approval monitoring and roll back. Takeaway five uh governance is how you stay in control while capability raise rises.

Uh if it can act it must be constrained monitor and govern it like a high power system. If it can act, it must be constrained, monitor and govern it like a high power system. All right. Um let me just add uh one angle that make people instantly smarter. Okay, there is a difference between model safety and system safety. Okay, let me repeat that again. There is a difference between model safety and a system safety. A safe model can be part of a unsafe system. Okay. A safe model can be part of a unsafe system because system safety include tool permissions, user incentive, business pressure, weak monitoring, poor robot

procedures, rushed deployment, data quality issues, human over trust. So that don't ask only is the safe. Ask if the entire system safe under pressure. That's the adult question. And if you build the governance layers we discussed tonight, you can deploy artificial intelligence, benefit from it, and stay in control. And that is the goal, staying in control. Okay. All right. All right. So, uh I don't know how long we we've been going here.

Um so now we're going to I'm going to just uh get ready and and jump into the demo crytography chapter three and chapter 4. We only covered yesterday neurotography chapter one and chapter two. We're going Let me just uh take this offline and get everything ready and have the videos ready for uh narrow cryptography chapter 3 and chapter 4 so you guys can get up to date into into that uh technology which is uh very nice. Uh yesterday we covered about neuro cryptography and I want for you to understand uh where is the future where we're connecting how your brain is utilized uh by corporations and the governments around the world and this is that is going to be like your Facebook, like your x.com, like your social media. This is the new frontier of technology. Uh forget about artificial intelligence. Artificial intelligence is just like uh distraction while all the things can be done. Uh more um in my opinion, more results, more measurable results.

Um and that's going to be the future. and how can we protect our brains from neuro cryptography. So it was a real pleasure. Thank you for watching today's live and if you're watching this later as a video uh thank you for supporting my content and my name is Victor Rosario. I am the owner and CEO of the Equation Agency and I'm also running for uh for mayor of Miami Day County. Uh so any questions just let me know. Okay, have a good one. Take care. Bye-bye.