Daniel Khachab

So this is where I disagree. I think it's the other way around. So I think AI is the perfect technology for traditional industries. Why? Because the problem in adoption is not that they think, oh, digital is unimportant. The problem in adoption is like, I need to learn something new. I have done this forever. I don't want to change. I'm going to need training. My people are not ready for it.

That is the problem in adoption. Versus now it's like, no, you don't need to learn anything. Do you people know how to use WhatsApp? Yes. Okay. It works like WhatsApp. type in what you want, it's going to give it back to you.

That is the problem in adoption. Versus now it's like, no, you don't need to learn anything. Do you people know how to use WhatsApp? Yes. Okay. It works like WhatsApp. type in what you want, it's going to give it back to you.

That is the problem in adoption. Versus now it's like, no, you don't need to learn anything. Do you people know how to use WhatsApp? Yes. Okay. It works like WhatsApp. type in what you want, it's going to give it back to you.

So the adoption curve is going to be way faster, like AI is the perfect tool for traditional industries, much more actually for startups and tech because those are people that are tech first, they know how to work with interface and stuff like that.

So the adoption curve is going to be way faster, like AI is the perfect tool for traditional industries, much more actually for startups and tech because those are people that are tech first, they know how to work with interface and stuff like that.

So the adoption curve is going to be way faster, like AI is the perfect tool for traditional industries, much more actually for startups and tech because those are people that are tech first, they know how to work with interface and stuff like that.

Yeah. I don't think everything needs to be in LLM and not everything needs to come off the cloud. I think for many use cases, small language models are completely sufficient and many of which can be hosted on... Even large language models, they can be hosted on-premise. Nothing's going to leave your doors. So you can create a product... with this as well.

Yeah. I don't think everything needs to be in LLM and not everything needs to come off the cloud. I think for many use cases, small language models are completely sufficient and many of which can be hosted on... Even large language models, they can be hosted on-premise. Nothing's going to leave your doors. So you can create a product... with this as well.

Yeah. I don't think everything needs to be in LLM and not everything needs to come off the cloud. I think for many use cases, small language models are completely sufficient and many of which can be hosted on... Even large language models, they can be hosted on-premise. Nothing's going to leave your doors. So you can create a product... with this as well.

And it's completely hosted even on your own internal service. And so I think you've got to go through the same questions around data security that any SaaS has to go through. So that's maybe a one-on-one, but then on the adoption curve, you still win.

And it's completely hosted even on your own internal service. And so I think you've got to go through the same questions around data security that any SaaS has to go through. So that's maybe a one-on-one, but then on the adoption curve, you still win.

And it's completely hosted even on your own internal service. And so I think you've got to go through the same questions around data security that any SaaS has to go through. So that's maybe a one-on-one, but then on the adoption curve, you still win.

So I do think that actually two interfaces will survive. One is kind of like the interface in which you ask the AI to do something and which it returns you what you want. That can be, you know, how much revenue did you make last month? It gives you back a number, you know, make the payroll that gives you back the result and things like this. But AI is not...

So I do think that actually two interfaces will survive. One is kind of like the interface in which you ask the AI to do something and which it returns you what you want. That can be, you know, how much revenue did you make last month? It gives you back a number, you know, make the payroll that gives you back the result and things like this. But AI is not...

So I do think that actually two interfaces will survive. One is kind of like the interface in which you ask the AI to do something and which it returns you what you want. That can be, you know, how much revenue did you make last month? It gives you back a number, you know, make the payroll that gives you back the result and things like this. But AI is not...

It's not God, like it makes mistakes, just an intelligence. It doesn't have superhuman knowledge in particular, not on your proprietary data. And so you got to train it. And so that's the second interface that I think we will have in the future. And I think that we need to think in a way that, hey, How do we make it as easy as possible for our users to train the AI? That's the key.

It's not God, like it makes mistakes, just an intelligence. It doesn't have superhuman knowledge in particular, not on your proprietary data. And so you got to train it. And so that's the second interface that I think we will have in the future. And I think that we need to think in a way that, hey, How do we make it as easy as possible for our users to train the AI? That's the key.

It's not God, like it makes mistakes, just an intelligence. It doesn't have superhuman knowledge in particular, not on your proprietary data. And so you got to train it. And so that's the second interface that I think we will have in the future. And I think that we need to think in a way that, hey, How do we make it as easy as possible for our users to train the AI? That's the key.

It's not going to be everything's going to work from day one, and that's not what you should expect, and you should not oversell on it. It's not like, what's the day one accuracy? It's actually the rate of learning that needs to be as steep as possible.