Why AI is not a search engine like Google

So let’s let’s get into it directly. First thing, this is a very basic. I just want it out of the way. AI is not replacement of Google. Because a lot of people say, “Oh, you don’t need Google anymore.” It’s nothing to do with Google. It’s not a search engine. So if that’s what you’re thinking at any point, please stop thinking that in any way. Why? Google has a huge database of data. when you search tries to match your query and sends you data back. AI has no database of data. AI has no database of there is no data behind it. These are just numbers, probabilities which are saved and it tries to match your query and it is answering it’s creating that answer.

That’s why people talk about it’s a completion engine and other things. It’s creating those answers. It’s not a search. It’s much more powerful. It’s creative. can do a lot more because it’s powerful. It’s dangerous also and it can go crazy. We have all seen that, right? And people keep coming up with examples of weird answers which you get from AI and still happens will keep happening. Um and and so that’s one thing I think we all know that by now, right? We’re clear.

Can I ask you a question? So you say if I go into one of the uh goo one of the AI and I ask it a question y like I would ask Google you’re saying it does not go and find the answer somewhere out there. No it does not in fact if you like as experiment later I will show you I will disconnect Wi-Fi. I have a version of model here running locally and it runs on my laptop. This is a ThinkPad with 96 GB uh uh you don’t need 96 64 will work and and just one standard everything is here you can ask all the questions it’ll answer everything from this but how does if I’m asking it specifically I ask it for a state law yes in Canada uhhuh how does AI know find the is it you saying it sometimes makes it up it absolutely makes it up.

So I’ll give you example so so now now now let me preface it they have gotten smarter So if you ask it, you can ask AI now, you can ask GPD for example, you can ask, hey, what is the Austin temperature right now? Now what does it do with that, right? Used to be it used to make even that up. Now they got smarter with that. So now their inference engine on in front says there is no way we could have have this data because they’re asking something current or stock price today, right? Dow Jones index, what is it today? what it does it calls a tool for web search in case of open AI they were using bing I don’t know what they’re using now they behind the scenes they do the search bring the data and the report so now’s AI there is a part of that search and in fact if you use depending on the models you use which show the thinking like deepseek and all they actually show you they will show I’m going to the web now I’m going to find this data so when it is connected it does but the AI model makes it up.

And examples I’ll tell you which we which we dealt with is so we have quite a few marketing companies one of them is Kou which is the one which we got to AI but we have true blue collar and few others so I asked AI hey tell me what kuare does and it answered it really well same thing the way you’re thinking this is three years back right and I said oh freaking great it knows what ku does how does AI know it knows okay it must be somehow getting the I said tell me about true blue collar True blue collar. Oh, true blue collar. It’s a organization for blue collar workers. Probably not probably. It said it’s a organization for blue collar blue collar workers. Said, “Huh? It’s actually a marketing agency. Other agency I own.” But we started that agency a year back. They didn’t have that data. So it made it up. It made it up confidently.

Then I said, “This is not true. Did you check the website?” It did not because that time the website connection was not there. So it will so what it has been trained to help and it’s the help is the best way I describe this this is actually important we should spend some time here that when you how to deal with AI AI whenever you’re thinking of AI think of a human being a good smart helper human being when they don’t know what do they do they make it up we make it up every day right or whatever level we know we think this is enough now most of us are smart enough not to make up to the extreme and AI is getting smart also that let’s not make it up so much that it’ll get so the newer one they’ve trained it but small subtle things it still makes it up and it doesn’t tell you because we do every day.

If I ask the same question of all of you if it’s got a little bit of a philosophical element or little bit of a uh uh variability every one of you ask the same question will answer it differently right because your training is different it’s the same way for LLM there is a variability so it does make it up and but it’s that’s the definition of this model is generative. It’s supposed to make it up, right? But it’s supposed to make up mostly right answers, right?

So So are you able to ask it if it’s made up? Yes. Now the modern one you can earlier it would actually lie and say no, I found it. It’ll actually now it’ll say it will tell you because they figured that out and they added it. It’ll say oh I could not get internet. Sorry, I didn’t have internet this time, so I just made it up or I thought it must be that.

The other one it used to do is give it a YouTube video and say, “What is this video about? Can you just analyze?” And everybody wants to analyze. I said, “Oh, great. This is it analyzes video and tells me.” I said, “No, that video has nothing to do with that.” Oh, but the title was that we could it could only read the URL title. It did not read the because then it could not. But now Gemini and others can actually read u video. So it’ll make it up because that’s how it has been trained. So they’re adding few more things to help tell you if it is make it up.

But it will always make up the best possible answer to help you. That’s something to keep in mind. So if you want the right answer, you go to Grock and a couple others and see if you can absolutely answer there is something which I use. It is from Andrew Karapati.

Using multiple AI models to compare answers: Council of LLMs

It’s called Council of LLMs. So what I do is when I have to really find out things, I ask a question. So this is a wrapper around multiple models and I ask the question that question goes to all and there is a chairman who says hey this person is saying this this person is saying this this person is saying this why are they argue amongst yourselves it shows the arguments going back and forth that this one is saying this this one is saying this one is saying this and and I I’ll I’ll share the eventually if you give me email list or something I’ll share the link to this this we have we we have built this already anybody can use it kind of a thing and there is a free tier.

So, and it’ll basically show you all of them arguing and then final answer will come out. That answer is always better always better because you can see exactly what happened because they all have different strengths. And so, let’s talk about that. Never mind the presentation. This is useful information for everybody.

The strengths are chat GPT is a philosophical model. It’s great for general discussion of thinking about life and things like that. It’s really good at that. Grock is amazingly good at what currently people are talking about. Why? What is it based on? X, right? X is the data it is reading. So if you want to find out if did Oracle release this and in this new version this setting works something very specific, how to set up something in your uh Fusion, you want the new setting, does it work? All the other AIS will screw it up. Grock will get it right because somebody must have asked that question on X or Reddit or somewhere. They get it always right. Always right. Right.

What is Claude good for? Claude is like a button-down suited person probably finance guy. Always right. Wants to be exact. Wants to control things. Right. Makes up least amount of it’ll make it up but least amount because it’s very strict in answering. It’s it’s very uptight too in answering some. So it’s not very creative many times if you want creative answers.

So these are the strengths. So if you use all three, they’re going to work through and produce what you want. Right? So that’s that’s and and the other way you do this is with prompt thing when we because you guys are asking these questions. I’ll show you that why prompting matters. We heard right prompt engineers and there was this new task and everything. Uh and and it’s it’s actually not that hard.

I’ll keep reminding you. I’ll do it this many times. Whenever in doubt, think of the AI is a human being. If you asked a human what all they could do wrong or right and how would you ask a human if you do that you will get the prompt right? If somebody new walked in who’s never worked in your department, if you tell them do this, this this, you will tell them a lot more, right? But somebody who’s been working with you every day, you will just tell them, hey, get me that file and they will know. It’s the same thing. You’re working with the model, they don’t know you. You’ll tell them more. That’s a prompt, right? So that is what those things are.

Okay, let’s see if there’s other Thank you for love these questions because it I’m not going to go into what is AI.

What is AI?

Anything which definition is very vague basically the bottom line is what we used to think humans can do when computers can do that’s AI because you will get into a philosophical argument of what is it this is not AI even AI people argue that the current one is not even AI it’s making it up the AI is supposed to be like human brain and exact so let’s not even get into it and it’s based on it’s based on the original AI training systems and all even generative is based on what was happening in ‘ 80s and 70s and things were already there in place.

So, but currently we are in generative stage but anything which so that’s basically the definition you still use probabilities like the most accurate the next accurate and you can see generative generative AI does so this is this is this slide will get that clear so there are two um two AIs predictive

Predictive AI vs Generative AI

used to be the one which failed kind of not failed but in fact by the by the way uh your Oracle the finance part of the module or the accounting part of the module uses predictive majorly because it’s accurate in you cannot get you cannot guess numbers there right right so that part uses predictive part so there there’s lot of predictive in in in Oracle or it’ll be any financial system accounting system will have lot of predictive what is predictive predictive the data has to be in right boxes in the right format you got to tell me what this number is what field it belongs to which month it is then only I can calculate right

this side generative no you just throw all the spreadsheets data and everything don’t even define the columns it’ll try to make sense of it like human would you try to guess this column probably is this they didn’t name it but I think this is the this this is monthly probably because they have 12 months there oh they they didn’t call it month but there are 12 it can guess that one will try to guess this one unless you give it the 12 months data it will say I cannot draw the graph for 12 months it’s not possible that side will draw a graph it just make it up right it’ll try the best guess basically so these are the two so answering your question that there is a predictive side the big difference is this so the predictive AI

UPS saved $35 million annually by coming up with the predictive way of what route should they take the truck route thing they came up with the algorithm they should always turn right and things like that that’s all predictive it exact they did this 30 million 35 million they can save 4.4 4 trillion by going this route because this one cannot it’ll just stop I don’t have data I don’t know what to do that’s where actually humans are good right we when we don’t know we make it up actually that has value sometimes what we make it up think of uh science fiction it’s all made up right science fiction but science fiction has come out with ideas which turned out later possible it’s pure imagination pure imagination which turned out some things are possible some are not so if you can imagine early that’s what you do here. You can benefit from it. You can try that.

So this is not to be dismissed. The fact that it makes it up, that’s the big change in the AI world. This used to be looked down on. I mean like nobody wanted to even try this. But now they’re saying to the level where hey everything is actually kind of best guess anyway in the world why we are so stuck on this. Right? So this this this is what it is like as a business guy. This is example. So as a business guy, if I’m hiring a CPA, I’ll ask a CPA, what is 2 plus two? Everyone who says four, I’ve got engineers who can do that. The CPO who says, what do you want to be? That’s the CP I want, right? Right.

So so, so that’s that’s that’s because that’s the one who can strategically tell me few things which I can go. So that’s that side. That’s right. Right. So yeah I don’t know yeah so this has got you guys will have access to slides in case you want to look through but basically this is amazing but there’s other problem which is there this side you can go back track the path how 2 plus two was added to get four here it’s a black box you send the number in number came out how it was done nobody can tell even the people who build the AI models the generative models they cannot tell how we got this answer it’s it’s very similar to human brain how do we actually answer questions when we asked is it from memory I mean like in fact if I say 2 plus 2 I think it is from memory right nobody’s adding right but if I ask you to give me a square root of 63,422 you’re going to probably go through some steps so what is happening so which is it is it making up or what what’s going on we we don’t know about the brain same way about LLM we don’t know what how it does it I mean it’s set up so they know the structure but how did it arrive to that answer cannot be repeated it’s not known because the randomness so that’s called temperature there is a parameter which you can set with each of these models so the parameter goes from 0 to one you can decide how much probabilistic you want it to be depending on your job you can so if you’re really coding with it you change those but most models we use they’re already set at a number Okay,

The core competence of AI models

most of the uh slides I have here, they’re kind of in uh like infographics intent. So they’re dense lot of stuff on them, right? And it’s kind of intentional because instead of having one one slide, I just want you guys to get the big concept and there is some of them have repeats intentionally.

This is a foundational slide I would say about current now everything I’m talking about is generative AI not the predictive predictive is almost like math it’s computer it’s like get it right there’s not much as AI to learn you give it the right data it’ll always produce right great is that AI in fact I have had major arguments with the AI itself saying that’s freaking computer program right it’s a if then else basically no actually there is a learning in that too the fact it can learn from the data and start building that’s why it is AI so it is actually AI so there is so basically it’s you give it enough data it learns how to do the calculation that’s it figures Watson model the Watson IBM yeah yeah so yeah so they that’s that’s what they’re doing basically they’ve got logical path they follow they will playing chess they will get compute every possible scenario right that’s what they were doing every possible scenario and if they can do fast enough then they can beat a human it was a speed to how fast the hardware can run basically right okay so um so

what AI can what can generative AI can do if you get that then it’s rest is just doesn’t matter what can it really do so currently there are four things it can do understanding and four kind of media it can understand now text I think everybody understands that or you can type natural language processing and it can understand what you’re saying nowadays much better than earlier days when you used to talk to computer. It can actually understand images and this is understand means understand. It can say what is the object in there tell me you know this is used u in driving and others actually there is understanding video in driving but image is the basis of that right so it can understand image it can understand audio so what does understand mean understand means you can give now a image to it which could be a spreadsheet printed and ask it to interpretate it earlier you had to give the spreadsheet as a data sheet you could not get a screenshot and say analyze it because because it can actually understand the image. It can read from there, right? It’s this is more than OCR by the way. It’s just not OCR. OCR is part of it. Everybody knows OCR, right? Yeah. So, so so it’s a lot more than that.

So going to uh on the generating side, it can generate text. So once it’s given the idea, do this for me, create this video for me, create this, it can generate images, audio and video. It generates amazing music. By the way, that’s the part I have started wondering that we kept on calling this paintings and others in art. There’s a whole question that is it what is art? How do you define it? Why? I mean like why one is better than the other, right? You know, Matise was amazing. But if nobody told you that this and just somebody made up something from AI and would you know difference between Matis and AI? Probably not. I’m not talking about just copying Matise or or artist actually coming up with a line of design, right? So it’s very subjective artist. So it can do it right.

There’s a display at the Mo the Mona right now the Mona modern museum of modern art in New York City. It’s a 24 foot by 20 foot AI generated piece of art. Yeah. That changes depending on who’s watching it. And they’re saying the average person looks at a typical exhibit in the museum for about 30 seconds. And this one they spent 18 minutes looking at. because it’s just fascinating actually I can already I don’t know about that but here is the one which being on the dark side I know if I create image which has somehow likeliness to you you probably see it you’ll watch it longer right I mean just one example of things you can do with AI you know because having a camera figuring out who you are and immediately changing things absolutely okay so and transforming transforming kind of you know it can translate it can take one video and translate to the other so any any transformation

But if you notice it’s intentional I repeated these text images audio video. So if nothing else if you can remember AI can see it can transform it can create right and it can reason with that data. That answers all your questions about what can AI do for me. So all you have to now look at it oh I need AI to figure out best way to place these objects here or you know the table design the best. It can see right how do we do it? We look at it and say oh this looks cluttered right there should be a better way to wire this oh yeah I can do that now for you so all you have to do is go and say hey this looks cluttered can you come up with a better design because it can see just upload the picture so that’s if you remember just that it can see read write and create and these are the things it can do then you can ask it I’m not talking about just chat GPT if you’re building an app or you’re asking Oracle to add things to your apps so you look at the business which I’ll cover a little bit that people should not worry so much the FOMO of we I’ll get left behind.

Uh that’s very common for everybody including developers and everybody. It’s not don’t worry about that. It’s just a tool. The problem is like as a human you got to figure out what is the problem in the business. What are you facing? What is the thing you want to solve that still you have to figure it out right? You could even use AI to analyze and say these are the things I’m facing which one is a bigger problem but you got to still identify those problems that’s not going away. You need human beings to do that and then AI will help you with that. So that part has remained and will always remain. So look at the business.

So in fact the people who look at the business who do lesser I have a recently a post uh on LinkedIn I talked about there was a professor Dystra. He is known for shortest path algorithm. He never used computer. It’s one of the most known computer science algorithm. He did not use computer. He just came up with it. Concepts that matters more than AI would have not come up with that. Right? So those thinking like thinking what is the problem? What do I want to solve? Concepts still matter. asking the right question matters even more now right than ever before. So as long as you see it that way there’s nothing to worry.

Other other reason there is for more is I think is these all these AI companies they spend so much capex is so high they want all these $20 per month users. So they tell everybody if you’re not using AI you’re going to be left behind. You’re going to be left behind. So just subscribe subscribe pay pay pay. Right? They scare everybody about this. So even though I mean like it’s it’s counter to I’m in AI business. I sell AI. I like that. They scare everybody. Every company is scared. They come to us and say fix it. Right? But I usually tell them first find out.

And that’s other problem is people don’t realize that sometimes we talk to business owners and they’ll tell me oh this this this I’m having this problem. Our customers hate how we do this this product or service. Say that’s not AI problem. That’s a business problem. It’s there’s nothing AI can do for that. You have a mismatch. If you have a wrong product, right? What can AI do for you? So if you have a bad business, AI is not going to solve it for you, right?

Okay, this is the other general AI concept which applies to everything useful. So the kinds of models and this

AI models spectrum: Access, Cost, Privacy and Control

is actually very relevant the special the middle ones and right one is very re relevant to Oracle and uh Oracle CL especially the Oracle cloud uh that’s what Oracle is pushing for big time. All these models the big ones we know they are in this category which is u you know the open AIS GPT4 now five Gemini uh these are all closed models it’s like a taxi you pay for the cab you go away there’s nothing you can change it is as it is it’ll get you from one point to the other but you can’t control what it can do it can do don’t expect to change much

the open source models and uh guess which country is way ahead on China they have done everything open source right and and I’m surprised that US has not learned from Linux and others open source is a good thing right it does get the innovation goes faster much much much faster one other thing by the way that’s my personal opinion uh and Microsoft had considered it but they did not make they should have made Windows XP open source once they had moved on to the next Windows they were thinking XP was very popular probably the most popular operating system they did uh those of you who’ve been around probably know uh it was very stable system and when they moved on to the next one they had considered let’s make this open source if they had Linux would have died people would have just installed Windows XP everywhere for free and and and that would have taken off but they just never did enough right and Linux has taken over all the server world

um the so the so that’s the open source but really this open source most of them they’re not fully open all they’re giving you is what you’re what you’re talking probability The one way to describe is sometimes they use as weights just the model structure and the weights. You can change the weights. Weights is what defines the training. So you can change the training but you cannot change the structure. They don’t give you the training data. True open source does exist. They give you even the training data. So you can actually repeat the experiment. You can retrain them or you can see you can do the same thing what they did. There are models in this space too. Very few but that exist. Most of it is here.

Why this is interesting to Oracle and a lot of companies if you want your data to stay with you or in your private cloud I’m not even talking about the local even if it is the Oracle cloud but it’s still you can get a p your own private version right so then this is the only way this way your data is going up

if you were talking to I don’t know if this is true I’m making this up but most likely true if you’re talking to chart GPT and or uh uh right now and you have we’re discussing something about Iran probably US government knows everything about you now. Yeah. Okay. probably cloud not that’s what they got upset they didn’t allow that right they say oh no we don’t look at your data but man your every query is going to their database right there this I mean if government orders I want to watch everything but in this no unless of course Oracle opens up their data center and gives it to government of course that can happen but you could also do your own local service cloud too right with this so this one is so guess what all corporate big clients are going to do they will be here not just for the government oversight they’re even worried about you know the compet compet competition getting access to their data right and what is happening with that so this is this is the play for Oracle other big companies who are providing the infrastructure to host your own stuff right so the open source models

so other way to think is that open source is you got a brain you can retrain it you can reshape it you can do comfortable with it at that point. on the other hand. But then how do we use we do use we do people do build applications with cloud you do code and others how do we do that you put a wrapper around that brain the main brain use when before you send the data to it you do something with it right and that’s that’s how you use those ones so here this is how this is done so prompts how you ask the question changes how you use the brain so prompt is one of the ways to do it context is any additional information you class. Then there is retrieval. Retrieval is if you ask for something and the data is in your database. In this case, I think Oracle fusion is perfect perfect example. You have all the different departments data.

If it was connected to it, even with this, it’s going to go and ask for that data, retrieve the data and then use the brains. But we are not able to change this, right? But if you had your own, you could have actually had the data in the brain. If you had your own brain, all the data could live in the brain. It will not live in the brain. It has to retrieve it then work on it from outside because you cannot change the brain. Right? And the tools are web search thing we were talking about.

Those are basically tools. These external things which allow you to do additional things which you need to do. It’s not model doing it. They’re just using standard stuff which was existing. Right? For stock market, they have access to the there are stock services which provide you the uh stock price. So they just call that. So that’s the tools you connect to it. Right. So that’s that’s how uh the existing models get used

Giving superpowers to your AI models

and and I think I kind of covered it. These are the terms you’ll hear. Skills is uh if you do something repeatedly you can teach it and the model will repeat it for you easily. Tools is external world and rag is what I was talking about the retrieval of data from external sources. Rag is basically a uh Oracle is to me is uh being the oldtime Oracle is just a database company. So that’s what it was always known for. So rag is vector database. It’s a yet different kind of a big huge database. It doesn’t have standard way. It just has numbers stored in a different way.

So you store all those. The reason why numbers because again we don’t store I mean this is one of the common uh uh examples that oh all GPT is doing its guessing. So if you say I love, it’s going to say you because it’s doing completion because that’s the most common sentence probably, right? But it is not storing I anywhere or love anywhere and you. So then what is it doing? It’s actually storing relationships as numbers and weights with no number. It’s not like it’s translating I to a number. There’s no So this is important.

LLMs do not show store any words anywhere but they still get it right because it’s just probability of which number goes next to which number and they convert it back to letters right it’s almost like the password this by the way it’s exactly same algorithm as password if you if you are in any good corporate system your system administrator cannot give you your password right they can reset it because that password is never stored anywhere the password is taken a hash ing function runs on it, a new answer gets created. There’s no reverse way to go and calculate your password. It’s only one way. That’s how LLM search works.

When you ask for a question, it converts it to again like password to the language which is stored and matches it with that. It’s never matching word to word. So there’s no backward way to go. Good. I’m sorry. I just keep getting a little techy in there. I love the stuff how it works. But so then what about when you ask like the next question? So it says like you know do you want to know more right and then you ask a deeper question it’s still going back to what it just brought down.

That’s the context. Okay. So the what it uh what other thing they do is every time you ask in a thread a question whatever you asked easiest way to think of it is it sends the question and answer back to LLM. It says okay you know where is uh uh what is the weather in Austin like? Okay. And then it might say are if it says uh uh uh like how is the uh is the airport busy today? Then it might those two questions if you ask them in a sequence when you ask the third question it sends the question and answer last two back to LLM. So if you ask a question, it can make connections like human because it knows you asked those two questions. That’s what it is doing. So the memory use keeps going up and up and up. But if it hallucinated on the first one, then so hallucination works off of that.

There is a context window which is getting better and better. It used to be 250 case 1 million number of tokens which is think of number of words. So what it does is it’ll keep sending, keep sending, keep sending and then it can do two things once it runs out of those tokens because it cannot send. The total it can send is only 1 million. Say what it’ll do is it’ll stop dropping. It’ll start dropping from the top. That’s kind of a it’s it’s it’s a moving window. In a moving window, it’ll just forget the past.

Or in Claude and others, what they do is when they start reaching around 80%, they take everything else from there and they summarize it. So you lose something but the sum will remain and it keeps summarizing and that if you do do that summary of a summary of a summary eventually it’ll elucidate but in the moving window it’ll start hallucinating much earlier and you can notice it man I told you that why you were asking me again I already told you that right because it lost it already but if in the summarization thing it’s kind of eventually that summary of summary of summary will lose something right that’s how that part is done right questions

I’m going to jump to uh you guys on this here. This image done just for you. So uh there’s a good news for enterprise SAS. Why?

In enterprise SAS, the big deal about enterprise SAS is you have data from all these different departments which needs to be analyzed and it’s kind of a hard problem to find patterns and everything. Guess what? AI is amazingly good at that without you actually putting it in a spreadsheet and everything. It can make connections and it can find things and correlate, right? So that gives it it makes it really really good to do. And this is not the deterministic one.

I’m talking about the generative one. It can just look at the data from all different departments and AI on top and can give you some guidance what’s happening right and more you train it how to look for it can get better. So that’s great and that is what right now every company who is in enterprise space is selling hey RAI can do this for you. You don’t have to worry about that. You can now now if you earlier you had to go and ask a question go into a field and search for it. You can just talk to it. I’m going to give you a chat bot and you can just ask any question right

one other area I do not know how far it has gone but I do know this this Oracle model is what uh I was tell I keep telling team that we even as a marketing agency said we need Oracle model you sell them the software it’s so complex that they have to pay you month after month to run it right so so we want as a marketing do something for them but they have to come to you to keep doing it right So this case in this in this case um this idea so and and and the other part has been there are so many consultants right for SAP and others there are companies who just do nothing else but Oracle consulting and everything in fact um all my technical buddies from old days they just kind of first work for Oracle as consultants and then they leave Oracle and they are consultants for Oracle still but not directly because they can make more money directly right because why just setting up these systems was complex one area.

I do not know how far you guys are. AI is amazingly good at setting up things automatically. So instead of going through and filling forms, you could just chat and say, “Hey, what kind of departments do you have? How many this you have?” And it’ll probably configure it. So if not, it’s going to be there. A configurator for systems through AI just by talking to it and and enough of software is doing that now. Just talk to it and not worry about filling a form and going through and collecting. just keep going and it’ll keep asking questions and and it can do that. So that’s one area.

Analyzing the data configuration will become easier and easier with AI and analyzing the data will become easier and easier and those will be the things which Oracle and others will sell. Yep. So um you’re talking about data retrieval and data interpretation and using natural voice to to Yeah.

What about data input ensuring data accuracy? Abs. Absolutely. So even the data input is the same way right kind of where it was coming from depending on the sources or actually even even I I’ll instead I’ll go to the other area which is there used to be these middleware needed even now needed right you got data in this format and I want data in this format and lot of engineering companies make a lot of money just building that connection correct now AI can build that on fly it just it just give that data structure it and look at it and say oh yeah it’s like this and it can figure it out and I want it this way and I’m going to automatically build it on the fly and make the special connector just for this right so the data coming in can be lot easier with AI now for sure absolutely

right and even the companies who are doing those those middleware now if they’re using wipe coding or AI even if the AI doesn’t do it in your system automatically which I’m pretty sure Oracle will start doing that these now if it says right now I I we support five kinds of inputs it’s going to start saying oh we support these 30 kind of databases now because our connector can be generated on the fly if you give me the format of the data so that’ll happen if not then the third party companies who were charging so much more for connectors they’re going to charge way lesser now because this can be done very very quickly now I’m I’m talking about 10x faster 100x faster and and as reliably as the programmatic like middleware the the middleware the the end result is program the difference is that you had to define every rule for every structure AI can figure out that rule faster in fact programming is one area because if you think about it um when you’re training somebody on a data and you’re repeating and guessing what is the thing which repeats most it’s the code almost always looks the same right so it’s easiest to train on so it’s not a surprise that cloud code does really Well, in fact, that’s a one of the theory we we were talking about is people try to uh Next.js and others,

they want to write programs with the model languages. So, the theory of me old-timers we were saying is you know what if they just stick with the PHP and LAMP stack the older ones, you might be better off because it’s probably trained on a lot more old data than new data. There was more old data, more old code than new code. So your chances of getting it right in the old code is higher because all they did was they dumped the code and it looked at it and that’s what it’s trained on. So more so you know of course the new one can be done but coding is the kind of the easiest problem for AI.

And this might be a really dumb question technically. What is if you have if you surround your data with AI tools? Uhhuh. What function does the SAS solution play?

So the SAS solution so SAS solution to simplify SAS it’s more than this but the SAS solution what do they do? They basically have a common one interface where you can which produces the dashboard and reports for you and data is inputed in different formats and managed by each division separately. HR does their part, right? Salaries and all are numbers which accounting needs. The inventory and all are numbers which accounting needs. So from each one the data can flow. That’s what all these enterprise softwares are end of the day, right? That’s all they’re doing. You just it’s more complex than that. There’s detail in each, but that’s really what it is.

When you surround that with AI, what that function doesn’t go away that function still is needed. All it does is if the connection was not as good, AI can make it better. If certain things we could not read because the format mismatch, AI can quickly fix it for you. That’s one. The the the second one is analyzing which took humans to figure out the patterns. If you’re looking for a patterns, AI can do it better and faster. Those are the two things. It can really really help and give you some insights which with so much data.

Here’s a example of banking data. It’s not exactly this kind of a thing but banks love to would love to sell more products and services to their uh to their customers right if only they have so much data right they know what transactions how many credit cards somebody uses how many payments they make everything is there so if I was a bank if I see Mr. X has 15 credit cards and they are paying minimum which will be easy to see they are paying same amount I can send a personal email to them hey saying hey for instead of paying whatever rate you’re paying 10% I’ll take over all these payments and you have to pay only this much that’s so custom that person is going to sign up that’s possible right now but they don’t have enough human beings to look through everybody’s accounts to figure that out but AI can do it like this right and send that personal email automatically and have them sign up automatically and do everything. So anytime there is a big large data like this to be analyzed.

Could it be done before with programs? Yes. But would it have had somebody sit down come up with the whole logic? Now no. I could just describe this. I could go to AI talk and say hey look through everybody’s account and find if there are payments which are same kind and looks like credit card companies. Go and find the names of credit card companies so you can match the credit card. It’ll do. I can just talk like this. I mean and this is something I can do it with account right now. In half an hour I can build this and it will actually do it. In half an hour this will start working if it was a credit card as long as I knew this that this is what I want to do. So now like these there can be hundred rules it can build. That’s what can be done with the data analysis and all which could not be done before.

And quick clarifier is um AI we wrap uh software with AI. AI can reverse engineer all sorts of things and reproduce a system that’s better potentially. Yep. Here is answer. Bad news. Bad news. Yeah. So, bad news is this

Bad news for enterprise SaaS

that if I don’t know the numbers, I’ll be wrong because you guys will know better you that what size companies. But basically what happens is your software I mean yeah you can help me maybe what is the minimum size it’s like a $10 million company or $20 million company who who uses this software what is the smallest size company roughly it’s down to like 2 million 2 million right so $2 million company and higher end is biggest possible right fortune 100 companies right there is a point at which they don’t need you they can do faster custom software Now earlier it was not possible. It was too expensive.

That line will get drawn at some point because they might find that custom is why do they why they giving me this addition of modules and everything freaking I don’t even have that data right I don’t need that connector but because because it was built you’re charging for it. So two things will have to happen. SAS prices will have to drop. So it discourages people to build their own. This has to happen.

And enough companies will say, I want my custom solution. This is freaking too much. I’m paying too much to the consultant. I can just have one person write this for me and one support it. I’m done. That’s will be possible. And this is this is the problem. How far in the future? Right now. Yeah.

Right now. So I can So this is this is this is this is how I confident I am in I have not used Oracle product. If you give me just screenshot of every screen you have screens which is possible right somebody can say these are thousand screens if you give me thousand screens in two hours I’ll produce a demo which will look like it’s working and have all your screens working and connected in two hours the logic will not be right right now because but it will guess by the field what it is in two hours time the code will get written automatically with all your screens in Right now most people don’t need all those screens. Right? That’s the reality. So in a day or two I can have a reasonably working software. Will it have bugs? Absolutely it’ll have bugs. But now if depending on the money somebody will be willing to hire us and say write this man. I’ve been paying so long for so many years. I don’t want to anymore. So that line where will get drawn is different.

So that’s where one way is drop the price. It’s the same thing like make the music so cheap that people will just buy Apple right 99 cents who cares right everybody will nobody’s copying music the music copying stopped because the music became cheap right it’s easier to just pay Apple and have unlimited subscription so something like that will happen with the pricing with SAS every SAS will have to deal with this right if customers pay Oracle for innovation like custom AI will not create any innovation they can copy what you have Yes. No. No. Why innovation? Actually, I I don’t know. Uh who who know who knows who knows what they need more customers or Oracle? No. Oracle is trying to guess me as a customer. I’m going to say you know I wish this existed. Earlier Oracle doesn’t listen to me. I keep telling Oracle they don’t listen to me because they don’t have 100 customers who want it. Till it reaches 100 then I’ll do it. But now okay I’m going to sit in one week’s time I can have that feature right. I can customize my completely immediately that’s possible now.

So this is the yeah so that exists right I mean like so those are the kind of a things again I’m not saying is but there is a counterargument I’ve say is that hey just because you can grow rice doesn’t mean we all all grow rice right we still buy rice right just because it’s somebody comes up with magically faster way to grow rice in your backyard doesn’t mean we’ll start growing rice so there is a point right so there’s a line which will get drawn not everybody will because just because you can build software and that’s the other part which is other extreme people say I can wipe code anything I can write my I don’t need any app I’m going to write yeah you can write your own word you can write your own why would you is that what you do for living like so there is a point right but if it is too expensive then enough will get written or a version will get written which will compete with you in our case we are a small company we have one of the software we have is uh we track because we have remote team in South America Philippines and India and US so we track time and we tell them hey we’re not worried about what you do but approximately get in 7 to N hours per day. We want average over the month.

So we use software called test time and it is I think $20 a month or so per user per user. So it adds up for 25 people, right? So one week we wrote our own and it is actually better and now we thinking of selling it for $5 a user to everybody. Right. Right. So it’s it’s is that and did we we didn’t even know how to start writing that software. Our coders did not know. We didn’t hire anybody new. It’s the same developers who do general developers. Now they could do it.

H how did you how did you start it? Did you do the the screenshots? No in this case. Yeah. Screenshots later because it was more like first we did the functions part. These are these guys are coders. They didn’t start with a screenshot. They started with the functionality. What I want? I want the time tracking. Just put it in a database. In fact our our guys are hard coders. So they don’t even care about the screen. They said let me get the data in database.

I’ll check everything is working. So just collect when somebody logged in, how long and everything and just start saving that and and to do that you just went to the they went to cloud code and started talking hey I want to build something and it needs to work on Windows it needs to track. So cloud code said oh yeah there is this service in Windows which tells you which program is running. So you can call that and so you do this discussion it tells you what all earlier you have to search through Reddit thread find some expert which service should I call how do I know which program is running now it’s like you’re just talking and you’re getting the answers back and back back right

you mentioned claude a lot you seem to like that one yes uh for coding yes I love for philosophical discussions I love uh uh chat gpt uh and uh but claude for coding absolutely yeah and and and grock for finding out what’s happening currently Yeah, I mean Grock has its own biases and all. I mean, I don’t know what Alon does with the data on top. But when it comes to technical, it doesn’t matter, right? Because he doesn’t mess with the technical data, right? I mean, one of the interesting questions is ask try to ask uh any of these which I’ve tried very hard, they don’t answer. Hey, is there is always a conflict between science and spirituality.

Doesn’t matter what religion. So, logically, it makes no spiritually makes no to me. Spirituality makes no sense. There’s no logic in it. So can you help me when it comes to these hard facts when they claim this this this they say oh yeah kind of depend I mean it’ll always answer kind of depends you could be they will never answer they will so this this is the guidel guidelines or things but if you can take a Chinese model without the rappers real AI they’ll start answering they’ll answer questions because ours have put rappers around them don’t say this you’re not don’t do this is that a concern of yours the Chinese versions no I I I like Chinese versions Except they also if you use them directly they have there is a state in India called Arun Arunachal Pradesh and that state China claims that it belongs to them.

So if you ask tell me about Arunachal Pradesh the Chinese model says I’m sorry I cannot talk about that that’s all the answer it so obviously they’ve done something yes right to the model but any other model if you ask they will tell you that’s Indian state which is in China claims theirs and things like that right so all right so that is the risk here because software can be written quickly so there is a line which will get drawn right now it was the barrier to entry was so much lower it’s getting lower Now it was so much higher earlier. No, who’s going to build the software? But many companies will at price point.

Yeah. So AI um Oracle now kind of pivoting and talking about these things like our system of record for our financials for instance. Uhhuh. Do you think that and it has to be that has to be you know 100% right has to be unhackable. You know people can’t go in and steal the data. Is that still some area that is safe? Yeah. Yes. So then all the stuff that’s going on outside of that system of record could be AI driven. Sure.

So there is the other things generally security is a concern with AI but it’s it’s almost this is uh again it’ll show you guys my political leanings but basically when we say there’s so much pollution and we go against science who the heck told you pollution existed it’s a science and who will solve the problem it’s going to be science it’s not going to be some stretchy kitty feely thing right it’s the same thing with AI there is security but there are now programs which do nothing else but check security of the programs that’s one answer like so is AI secure is the first question second one is AI does have some weird problems for example if when you do authentication you log in and you’re checking and access levels are different of different users if you’re admin and accounting level you can see everybody’s salary others cannot right that’s easy in a standard system but now suppose you’re talking with voice and you you while talking you were asking questions you handed the phone to somebody else or you had somebody else on there I mean it’ll continue right there’s no more authentication so now what is the solution voice recognition actually tell a person changed I’m not going to answer that anymore and the data has to be siloed enough to know with the tag that this data can never be accessed by this right which is kind of done in databases right now but that that becomes even more important now so the challenges do increase because you’re opening the things. So there’s always a balance between openness versus closed. If you close everything, data doesn’t get shared. 9/11 problem, right? They had all the information but nobody was talking to each other. If they talk to each other, sometimes things leak. So there is a balance somewhere has to be found. It’s the same thing here. I think that

so yeah, this is by the way just for you guys. So basically I said just compare for me Oracle versus others. This is AI’s what a current AI understands in this case actually maybe let me qualify it’s Gemini the AI for this one is Gemini which I used. So it says that what I learned is out of that is that Oracle’s agentic part Oracle is way ahead than the others and that’s I think from what I know of AI that’s the right thing to do. Agentic is where AI can go auton autonomously and do some things and finish the work and come back and u in this Oracle is way ahead of others right and and uh so we didn’t even talk when we talk about the analysis and everything so think of this way that something which will run automatically behind the scenes and produce your reports find the problems in it and do everything for you behind the scene you never going and typing and asking so it’s not a prompt oh let me tell me this make sure the report is Right? Give me the 15 days report.

You don’t have to do all that. It did everything which was necessary behind the scenes. Oracle is ahead on that. I think if that’s based on this. So the reason this is relevant is if you see something here no our our team told us it’s the other way then all you need to do is like but that’s what is showing when people search. If somebody researching this is what they’re going to see, right? Because this is what it is. Outsider researching and figuring out this is what they get. So then if that is a perception, you got to fix it. It’s a marketing problem at that point, right? Because the data set or your white papers which are out there are just not right then or or I mean or they’re not prevalent enough or not visible enough to everybody to get the message out. This also has a uh actually uh there is a this is a propaganda war opportunity too.

You do it right. People who are searching I mean that was done always with Google. Now you got to do with AI. AI needs to start thinking what your company is and it’s going to present you that way and that will matter because people make those decisions now using AI because especially for complex decisions people do go to AI. If I want a comparison between two ERPs right now, it was very hard, right? Every vendor has their own slide deck and one looks always looks better than the other. I’m going to just take both the slide decks and put it in AI and say tell me and go outside and find out what people are saying also. This will happen for sure when you’re trying to sell now. So where would you go to ask that?

Okay, I’ll show you that. Uh how many people here have used notebook LM? Every one of you. So there are two or three exercises I’m going to give you. Every one of you at least twice a week or thrice a week should use notebook LM. Notebook LM is from Google. Um so if you have personal Google account you can get it from that and I’ll show you actually live. This is very very useful thing very this is from notebook LM. It’s the best research tool. I’m not going to go through this.

Business-first approach to AI

this is uh business leaders always think of the business problem first. Forget AI, forget AI, what is a business problem?

And many of you questions you are asking are business problems which is great. What will happen with this? How will the data come? Those are business problems.

If you keep thinking those then AI becomes just a tool and that’s how it should be. Don’t think of like oh AI you know uh first just think of the business first and it AI can solve a lot of these problems right growth problems efficiency problems inside problems uh for businesses and things like that and that is very very useful for a lot of times

right if you’ve been producing reports before just add upload that report and say do all this but I want the final report in this format it’s going to give you exactly in that format so that’s where it is very useful

and the other is if you give it three four examples of the format at then it starts saying what you you don’t even have to say what to do. You could sometimes upload upload four things and say here is the data I want something like this. Here are the four examples.

It’ll automatically create the prompt. You say okay you want something like this out of this. So it’ll automatically do the analysis because so that’s it because basically you’ve given it input and you’ve given example outputs so it can figure it out. So that’s the other framework.

Um I’m going to the other one I’m going to send you this in the email because we don’t have time to go through this but the problem is this is the first early discussions I used to have with AI.

I said your job was to be natural language processing and understand humans and not asking us to write like a computer you have to type the command find file and all but you have failed because now you’re right asking me to learn a yet another language which is called prompt you know it’s just a you just changed from one code to the other so you AI have failed basically right

you should have you should have gotten the sense right why you’re not getting it so I mean that was the initial thing and then AI guys answer you’re right it’ll get better but basically the thing is that the prompt why do you need to use this format why cannot I just talk to it and do the right thing right

so the closest you can get is what if you had a prompt writer if I gave you something you just talk to it and it’ll create the perfect prompt for you so that’s what we have created

I’m going to send you the instructions all it is is you can put it in chat GPT or you can put it in cloud code it has behind the scene it has put all the guidelines how the prompt should be written which have been done by researchers.

So when you talk he’ll ask you questions back and forth and it uses the guideline and writes the perfect prompt. Then you take that prompt use it. So you don’t have to write the prompt. So it’s a prompt writer basically.

So that’s what we use for all our work now. And it’ll be a much longer prompt of course but it is a really good prompt because it’s using researchers papers behind the scenes and the steps when you see the steps you’ll know you could do it too.

You could add more stuff but it is basically so that’s the so basically in front of instead of that long form calculation I’ll give you a calculator you know because prompt writing has become like oh there’s a prompt write prompt engineering prompt this you don’t need all that

you know with this so I’ll send this out uh this was if you had time we would have gone through the exercise and done this notebook LM so notebook LM is amazing and I’ll show you a demo and to prove to you that how good it is here’s the prompt you’ll get this slides.

Only do this if you’ve been talking to any of the AIS for AI for a while.

You can ask uh um ask AI this. Actually, it gets even more interesting if you’ve been spending a lot of time. Ask AI to roast you. It’s really good.

So the one one u mind which was interesting is I’ll share is it said AI you argue with yourself. you know the answer and you ask the AI then you double confirm and then you start questioning yourself so you are you know it already because I just kind of I’m the one who like I want to play devil’s advocate and all so the way it went is very interesting

and then initially when it first roast was not good I said no come on don’t be polite really do it then it it’s actually very interesting but it this only happens it been talking to AI for a while right

yeah this this is useful it starts giving you some insights this is a Very very interesting prompt. Again, if it knows you, it’ll give you a 90-day plan with the research behind it that why it thinks it’ll actually get you 50% happier and how will it track it?

Because I mean like first question, what does happy mean? Oh, there is actually research out there how to measure happiness. It’ll tell you and why this will work and it tells you 90 days and it’ll also tell you why it will not work. What are the things you might screw up? So, how to watch for that if you want to do this.

So, so um yeah, this is a brainstorming part partner prompt like if you are a business owner and you want to figure out it just kind of gives you a starting point, right?

This is where you hear I created a million dollar business. I asked AI to do this. Most of them are lying by the way. It’s not possible to create AI 1 million business by just kind of prompting but it can give you some guidance on what will work, what will not work right with that.

So that’s that’s that kind of prompt.

Um so I’m going to show notebook LM live next and here is uh things by the way uh which you if you guys need that newsletter is once a week I send out to business leaders which has insights on a higher level.

My newsletter is not oh here is a trick you can use because there are so many Instagram people who will tell you the tricks on this thing. they’re much more what will happen, what we think, what matters. Uh because it’s the business leader kind of a thing.

And on social media also, I’m new still there. I hate doing it, but I am trying to be there a little bit. Yeah. Because I’ll rather be sitting behind a computer and code and change things.

So I want to show you guys notebook LM because if nothing else, if you start doing that, it’ll be amazing.

So let’s let’s come up with something. What?

So this is a research tool mainly or a learning tool. So if there was something which you wanted to learn the big advantage of this one is it bases the data on what was input in it. Either you can upload all the files here then it’ll just give you the answer based on that no hallucination because it’s only the data you gave. That’s one advantage.

And otherwise if you’re just researching so I mean and the ones I’ve done. So we I want to do a live example because otherwise I have already I will show you how I did the ones which you guys saw the presentation.

So Oracle fusion so this is 10 sources which I used and it produced all this for me uh the data. So but it’s the big power is it can do it right here on the fly.

So throw something out some idea out which you want to know about some topic or some com maybe maybe your competition some you’ve been curious about who you think yeah know they keep running into and they are pretty good and sometimes a problem maybe that just do that yeah a question about workday

okay so I’m just saying this

I’m saying workday versus Oracle fusion how is that right so and it this is what it’ll and it’ll first because I didn’t here I I’m starting from zero.

Now if you had some reports or powerpoints I would have you could upload here everything Google docs and everything just throw it in. But in this case because I didn’t have anything I said just search and I’m going to just go stupid and say okay whatever you found just use it.

So it’s already analyzed all that data for me. It can do for me things like mind map from this data.

So I could have said in the mind map put something but in this case it is just the topic is so simple. So I’m keeping it like this and I can do this here in the slide deck and I’m going to say you know when producing this report

create in this format the comparison slides and I can create a audio overview and I can create a video overview. all will be not ready but whatever will be ready it’ll be worth looking at it

in this case I’m not even do the colors I’m going to just say just do it because you guys know that colors can be done and the infographic is probably worth it

so it’s it’s generating here right now so it doesn’t take too long it actually creates that audio one even if it’s not done later I’ll send you it creates actually a podcast of two people talking about these two and it’s like The quality is like and it has that um like what about that as if it’s real people talking and we can just post the podcast and you know get this

and then it the video of course the presentation it does and uh everything like it’s it’s it’s amazing what it can do basically on the F.