Gitau: DeepSeek's Simplicity Optimal for Africa

Bloomberg

Watch Gitau: DeepSeek's Simplicity Optimal for Africa - Bloomberg on Dailymotion

Transcript

00:00I wanted to just start here. When you're looking at an AI partner for your firm,

00:05what is it that you're seeking? And maybe what is it that Western providers give you

00:09that Chinese providers don't or vice versa? Okay, thank you so much for having me. When

00:16you think about AI, we look at AI from four different angles. In terms of like, do we have

00:22enough data sets? Do we have compute? Do we have the skill set and talent to do it? And do we have

00:28a market to launch our products into? And of course, there's a huge governance angle that we look at

00:34it from. And when we're looking at a partner, we are looking at who is the best to optimize across

00:39these five pillars, right? Can we have enough talent to be able to, is it a talent to be able to train

00:45against the model? Is the cost of compute for accessing and training this model accessible

00:53and cheaper? Is it, from a data set point of view, is it able to understand our languages? Is it easier

01:01to train for our context and our culture? So those are the things that you're looking at when you're

01:06assessing which of the foundational models or frontier models are you going to go with? And

01:12you optimize for what works best for any of the use cases that you're building.

01:16Have you found that some of the models from China just work better for the continent and for what

01:24you're doing? They work better because they are better optimized for the emerging markets and this

01:32and the global south. And I mean, if you just dig a little bit on how DeepSeq was created, it was

01:37optimized for compute, for less compute use. And so you can imagine even it's like founding principles

01:46if you might, if you may, built around that is how do we enable people build under a constrained

01:52environment. So that's, that's the ethos of DeepSeq and you see it across the way they use it.

01:59However, in terms of like tokenization, they're still far. Everyone is still far. Even the Western

02:05models are still far. So we're still underrepresented, but when it comes to just cost of access and the

02:12way they build DeepSeq that you can run on very simple machines, you don't have to have like

02:17extremely expensive machines, makes it a bit easier and accessible for most people. Yeah.

02:23Do you see Silicon Valley models though, potentially changing in order to optimize and be better suited for

02:30emerging markets?

02:31I was hoping you're going to ask me this question because DeepSeq was launched earlier this year in

02:38January. And I mean, doing some of the work you were doing last year was very, very expensive.

02:44And coming this year, you can see OpenAI, Llama, even Gemini have launching an edge version of their

02:53model, a low parameter version of their model that you can train with low compute, which makes it easier.

03:00Number two, it has incentivized like the frontier model providers to say, how do we support the

03:06emerging Africa and other, and I'm thinking about India and Latin America in building on top of this.

03:14And there's a lot of provision of like better compute credit to help you get into your first MVP.

03:20So I'm seeing DeepSeq did us all a favor by doing this because it enabled other people to see us as

03:25viable market, which is what I love about this whole competition. Yeah.

03:31How do you think about data privacy concerns though?

03:36Oh, so I would say this, I'm from Africa. Yeah.

03:4160 to 70% of data centers in Africa are Huawei run, right? Yeah.

03:45And the remaining, all of us in the tech space are either on some cloud, Google, AWS, Azure, all of this.

03:58They're not residing on the African continent. So if I'm going to wear my African lens and look at data

04:02privacy, they all are the same. My data is not residing on the African continent, it's residing

04:08somewhere else, right? So we have to think from that point of view. It is not like I'm an American and my

04:15data is sitting in an AWS server in Texas. This is, I'm a Kenyan sitting in Nairobi and either my

04:21data set will sit in China or we sit in Texas. I have to choose where it's going to sit. So it is not

04:27a here, it's that argument. I don't find it aligning to what being an African means. There needs to be a

04:34better argument against Chinese model or Chinese cloud providers than just data.

04:40Do you worry, though, about, do you worry, do you worry about getting caught up, though,

04:45potentially in the trade tensions by choosing one side or another?

04:50I will say something. I think, I think the Moroccan, I was just watching the Moroccan

04:55minister talking about their relationship with China and the U.S. And many of the African

05:00presidents have said this, Africa will work with a partner that provides the best for Africa,

05:05right? And yes, these are, these are challenges. We're all afraid. We, we, I worked for both

05:11Google and Microsoft. I have great friends and we have great partnerships with some of those

05:15companies. So there's always that fear. But if we live from a point of fear, we'll never get,

05:20we'll never get anything done. So we, we live from a point of, we need to get Africa moving and

05:26work with a partner that supports, supports those priorities and is action oriented towards the

05:33betterment of Africa. Yeah.

Category

Transcript

Be the first to comment

Recommended