meta_pixel
Tapesearch Logo
Log in
TechCheck

New Chinese AI Model Threatens American AI Dominance 12/30/24

TechCheck

CNBC

Management, Cnbc, Tech, Faang, Investing, Business, Disruptors, Technology

4.566 Ratings

🗓️ 30 December 2024

⏱️ 8 minutes

🧾️ Download transcript

Summary

DeepSeek v3 is a new AI model developed by a Chinese lab, built in just 2 months with less-performant GPUs and for a fraction of what it cost Google, OpenAI and Meta to build theirs. It raises questions of America’s place in the AI race and whether shelling out hundreds of millions or even billions into building frontier AI models is still a worthy investment.

Transcript

Click on a timestamp to play from that location

0:00.0

As big tech is getting hammered in today's sell-off, CNBC's Magnificent Seven Index dropping more than 1%.

0:06.7

Well, there's a new emerging threat to mega-caps massive spending in America's dominance in the AI race.

0:14.1

Dear Jabosa digs into that for today's tech check.

0:16.4

Hey, Dee.

0:17.4

Hey, good morning, Leslie.

0:18.6

So here is a name that our audience may want to write down, DeepSeek. This is a new free open source AI model that beats the latest open AI and meta models on key benchmarks, and it was made for a fraction of a fraction of the cost. Now, it was trained by a Chinese research lab that used Nvidia H-800s. That's a lower performance version of the H-100 chips that are cheaper, more available, and tailored

0:42.5

for restricted markets like China.

0:44.7

Now, I've been testing it out this morning, and on the surface, it looks and acts just like

0:49.3

OpenAI's chat GPT.

0:51.0

And in fact, it actually thinks it is chat GPT.

0:54.0

When I asked, what model are you,

0:55.8

it answered, I'm an AI language model created by OpenAI, specifically based on the GPT for

1:01.1

architecture, suggesting that it was trained on chat GPT outputs, which, leaving aside terms of

1:07.4

service violations, it means that entirely new state-of-the-art models can be built on what is already out there.

1:13.7

In other words, OpenAI's moat may be shrinking.

1:17.1

If a model like Deep Sea can emerge with competitive performance, minimal cost, and reliance on existing outputs,

1:23.6

it signals a rapidly shrinking barrier to entry in AI development, challenging the

1:28.3

current dominance of industry leaders like OpenAI, based on technical tests designed to measure

1:33.7

its coding performance, deep seek outperformed other models, including Meta's Lama 3.1 and OpenAI's

1:40.7

GPT4. Oh, and by the way, those are the latest state-of-the-art models. And that led Andresge Kaparthi, a founding team member at OpenAI's GPT-4-O. And by the way, those are the latest state-of-the-art models.

1:44.9

And that led Andres Gaparthi, a founding team member at Open AI, to post DeepSeek making

1:50.0

it look easy today with an Open Frontier-grade LLM trained on a joke of a budget.

...

Please login to see the full transcript.

Disclaimer: The podcast and artwork embedded on this page are from CNBC, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of CNBC and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.

Copyright © Tapesearch 2026.