00:00 Speaker A
Bloomberg reporting Deep Seek is preparing for an IPO. They say may file as soon as this year, seeking to raise more funds in the private market ahead of that IPO. Maybe take a step back. First, just remind people, um all the excitement about Deep Seek. What is Deep Seek and why has it become such a big deal?
00:20 Speaker B
Yeah, I mean, Deep Seek is an open source uh AI model, right? They have different models in and of themselves, but uh they are the company is is deep seek. It was uh came out uh was it last year, the year before uh and knocked everyone’s socks off, right? with their R1 model. Basically saying, uh we were able to train this on uh low powered uh uh AI chips uh and we have, you know, this great kind of uh capability.
00:58 Speaker B
Everybody said, oh god, we don’t need Nvidia chips anymore. Uh abandoned ship. this is it. the AI, you know, kind of boom is over. Everyone can do this on on cheaper devices. Turns out not so much, right? Uh while the the cost of training may be lower or maybe going down uh for certain models, the larger models, you know, these these big frontier models, they still need a lot of heavy duty hardware to actually train. And then to run as well. That’s, you know, the older models, it’ll be easier to run them, but as as the models increase in size, increase in complexity, they still need high power processors to run. So that’s basically the the kind of long and short of how we got to Deep Seek.
01:54 Speaker B
The the bigger thing here though is that because it’s open source, a lot of uh developers have said, well, we don’t got to pay that much to use this. So why don’t we just use it, right? Instead of going with Anthropic or open AI and, you know, not for nothing, but uh Deep Seek has powerful models. It’s not, you know, Mythos like, you know, uh Anthropic’s, you know, super high-end AI that can, you know, hack into different pieces of software, but it is a good AI algorithm and an AI model. And so
02:35 Speaker B
there’s a lot of developers around the world that look at that and say, well, why wouldn’t we just use that? And so that’s the draw there. And there’s this uh dichotomy in uh AI right now between the closed models, those those mythosis, uh those Claudes and the the chaTBTs, uh sorry, GPT, you know, uh 5.4, uh 5.5, uh you look at Google and Gemini uh three, you know, those are closed models, uh and then as well as uh Meta’s new muse uh Spark, I believe 1.1 is their their most recent model.
03:13 Speaker B
Uh and then these open source models. And the the idea is, okay, well, these are the frontier models, these are, you know, maybe not necessarily uh Spark 1.1, but you look at the the mythos and and you look at the the GPT 5.5, those are the the frontier, right? You’re going to have to pay a lot of money to get access to that and you’re going to get great performance for whatever your needs are. But maybe if you don’t need the absolute best, if you don’t need a Ferrari, why not get a Camry?
