Subscribe to Our Newsletter

Success! Now Check Your Email

To complete Subscribe, click the confirmation link in your inbox. If it doesn’t arrive within 3 minutes, check your spam folder.

Ok, Thanks
This Chinese AI App has topped iPhone download charts
Image credit: Mojahid / Shutterstock

This Chinese AI App has topped iPhone download charts

DeepSeek is causing an upset following its AI model release.

Oyinebiladou Omemu profile image
by Oyinebiladou Omemu

Silicon Valley has long basked itself in conceit with OpenAI sitting pretty at the top of the Artificial Intelligence (AI) game. But there has been a turn on events with the Chinese AI startup DeepSeek causing an upset in the tech’s capital following its AI model release.

This new AI model, DeepSeek-R1, released last Monday is blowing up in Apple stores topping the charts, and that begs the question: how secure is America’s dominance in artificial intelligence?

The Chinese AI model, designed for advanced reasoning tasks uses a multi-stage training process combining reinforcement learning (RL), supervised fine-tuning (SFT), and a Mixture of Experts (MoE) architecture, which ensures enhanced reasoning capabilities and readability.

While companies like OpenAI, Meta, and Google poured billions into acquiring the most advanced Nvidia H100 chips and data to train their models, DeepSeek has taken a different route. DeepSeek-R1, backed by R1 and V3 models—both open sourced models of DeepSeek—was reportedly built for under $6 million using 2,000 Nvidia H800 chips—a downgraded version of the H100, courtesy of US sanctions. This has raised questions about whether massive budgets are a necessity for AI innovation.

In terms of performance, DeepSeek-R1 achieved 97.3% accuracy on MATH-500, slightly outperforming OpenAI o1 (96.4%). It also ranks in the 96.3rd percentile on Codeforces with an Elo rating of 2029, while OpenAI o1 slightly edges out with a rating of 2061. DeepSeek-R1 excels in creative writing and question-answering tasks, with an 87.6% win rate on AlpacaEval 2.0.

OpenAI Launches Strawberry AI, Pushing ChatGPT Closer to Human Intelligence
The preview version is now available to ChatGPT Plus and Team users, with Enterprise and Edu users set to gain access next week.

DeepSeek-R1 shows its work and reasoning when responding to user prompts. That’s a big deal if you want to understand how the AI thinks and not just what it thinks. One of its most enticing features, it is open source in contrast to proprietary models like OpenAI's.

The ripple effects of DeepSeek’s success are already being felt. Asian tech firms linked to DeepSeek, like Iflytek, saw their stocks surge, while companies reliant on Nvidia’s high-end chips experienced a dip.

The timing is particularly interesting, coming on the heels of OpenAI's joint venture with the federal government to invest $500 billion in AI infrastructure. But now, DeepSeek's move, operating under budget constraints and sanctions, raises uncomfortable questions.

How long DeepSeek will stay at the top is a question only time can tell.

Oyinebiladou Omemu profile image
by Oyinebiladou Omemu

Subscribe to Techloy.com

Get the latest information about companies, products, careers, and funding in the technology industry across emerging markets globally.

Success! Now Check Your Email

To complete Subscribe, click the confirmation link in your inbox. If it doesn’t arrive within 3 minutes, check your spam folder.

Ok, Thanks

Read More