Mamoon Hamid
speaker
146 appearances
1 recordings
1 series
first heard Oct 2024
last heard Oct 2024
Mamoon Hamid’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
Yeah, well, hats off to Klarna for undertaking this. I just remember the time when we built an internal CRM at Kleiner Perkins, and it sort of cost us many millions of dollars and then ongoing millions of dollars a year to just upkeep. And at some point we realized, like, there's a great tool called... We could just get Affinity. Exactly. We use Affinity. It's great, okay?
And, you know, like, we had four engineers on it, like, working... Why? Why? Exactly. Why? And now I think it's actually maybe slightly different. That's too broad of a brush to paint with is if you think of the people you'd hire for customer support as labor that you would spend money on. Now, how about you hire developers to do the labor work for you in the form of AI?
So I get the rationale perhaps that Sebastian has around doing that internally, but I also get like, there will be a company that's going to do it really well for you and you will have to pay for outcomes and you have to pay for the number of tickets resolved by that software or that AI. So the question is, are you not willing to pay that?
And in most cases, at the end of the day, you're like, you know, I should just pay Stripe two and a half percent, okay, rather than billing it all myself. So the question will be, well, how many companies will go down the path of Klarna? The other side of it is that Right now, we're in this area, you talked about sugar rush. We're in this era of you're doing a bunch of proof of concepts.
You're just trying out all this cool stuff that's come into existence in the last two years and seeing what can I do with it? And every CIO at every large company is spending real money doing POCs. In many cases, you realize like, you know, it's hard actually to build this custom thing inside POC.
And this reminds me of like 25 years ago during the internet, everyone's spending a lot of money internally to do things on the internet. And then you hired all these consultants from Razor Efficient, Sapient, and other companies that came in and tried to help you with the internet. And I think that same thing is happening right now. And history repeats itself.
Yeah. I think we will hopefully never talk to a customer support agent ever again. Like someone you call for the airlines, like, you know, help me with my flight and get upgraded or I need to change my seat or can you cancel it because I can't go. You know, the bank, things of that ilk that you still scratch your head, why am I still doing this?
And it will happen, you know, hopefully just like you provide a text message and it gets resolved on the back end and it's all being done by agents talking to agents and stuff like that happening. I hopefully think the world will have a way better customer support experience. Please hold.
You know, I was on a call yesterday with one of my CEOs and he said, hey, can you hold on for a second? I said, sure. I got hold music from my CEO. You got hold music? From one of my CEOs.
There's a lot of time being spent on a lot of the middle layer between the foundation models and the applications. And I think in the middle layers, middleware, things that allow you to use those models better, faster, cheaper, and build applications on top. So if you think the application at the top of the pyramid, the foundation model at the bottom of the pyramid, you know, pyramid.
In the middle, you've got middle layer. There's a lot of new technologies emerging that allowed it, for example, to capture vector databases or waits for fine tuning of models inside of vector databases and things of that ilk. There's just a lot going on there. At the same time, some of the value seems fleeting in nature.
I think in early innings, when things are in such high degree of change, so the rate of change is so high, there is a lot of that investing that happens. I feel like perhaps it's a little over invested.
I don't. It's a bit like the hyperscalers thinking they can do everything and they've decided that that's a great business model is to own the electricity and then just charge for by the hour or the kilowatt hour. That's a pretty darn good business model for the hyperscalers that provide the models and opening eye.
I was at opening eye maybe a month ago and we're going through all these demos of cool products that are coming out like O1 and Strawberry. and realize that their positioning is we can't do everything. We're a 1600 person company. We can't build the application layer stuff that we want you guys to build or your companies to build.
And so there's a great business to be had in LLMs and in providing the compute and the electricity. And there's a great business to be had by being very vertically focused around applications.
The beautiful thing about just the GPUs getting better and the infrastructure being more performant, models getting better. Sure, the models are getting bigger too at the same time.
So let's maybe draw the difference between there's all the folks who are providing those GPUs, there's people providing data centers, there's people who've now built LLMs on top of all this compute infrastructure that's there. So clearly, NVIDIA is a great business. Hyperscalers are investing today for the future.
And I think ultimately the margins, just like if you look at 20 years later of AWS, how great of a business that is, that standalone basis would be a top four enterprise software company. Same with Google Cloud. So that's a great business over time. And then you look at the LLMs. So if you're just providing tokens, is that a great business?
Right now, given the public profile of OpenAI and financials that we've all seen, today it's not a great business, but they're smart enough to figure out how they can get to a gross margin that will allow them to be a highly profitable business over time. Do you think the scaling laws will continue?
Yeah, so just the rough math is that in the last 18 months, the price of a token has gone down by 200x. But that's just the super early innings, right? We're talking about the first two years ago, we didn't have chat GPT. Today, we have so many different applications that are utilizing this technology. So will it go down 200x over the next two years? I don't know.
Showing 21–40 of 146 · page 2 of 8
← Previous
Next →