Featured

TechDogs-"DeepSeek's Disruption Fake With $1.6 Billion Spend, Safety Test Fails, & Gov Bans"

FeaturedEmerging Technology

DeepSeek's Disruption Fake With $1.6 Billion Spend, Safety Test Fails, & Gov Bans

By TechDogs Bureau

TD NewsDesk

Updated on Tue, Feb 4, 2025

Overall Rating
In the last fortnight, the artificial intelligence (AI) sector bore witness to one of the most exceptional products it has seen in a long while, probably one since the launch of ChatGPT, DeepSeek’s R1 model.

DeepSeek claimed its R1 model was at par with OpenAI’s latest generative artificial intelligence (GenAI) models, that was until OpenAI launched a newer model among other moves.

However, the catch with DeepSeek’s revolutionary new model wasn’t just that it was at par with the most capable models, but that it was developed using a fraction of the cost, resources, and time that models built by leaders such as OpenAI, Anthropic, Google, Meta, and others used.

Reportedly, it cost DeepSeek only $5.6 million, less-powerful GPUs (graphics processing units), and around two months to produce its prized possession. Moreover, DeepSeek’s model was completely open, meaning that anyone could download it and use it as they wanted.

This caused a major swing in users migrating to DeepSeek’s new powerful tools, ditching those developed by the so-called leaders.

However, as time went on, numerous issues began popping up with the claims by DeepSeek.

As per a report by SemiAnalysis, DeepSeek misrepresented facts about its most admirable characteristics.

For one, the $5.6 million cost (said to be $6 million) associated with the development of the tool was true, but it only represented the GPU cost of the pre-training run. DeepSeek actually incurred expenses of $1.6 billion in sourcing hardware for its servers, along with around $944 million in operating costs.

Next, the money spent on acquiring hardware included sourcing 50,000 NVIDIA Hopper GPUs, which include 30,000 H20s, 10,000 H800s, and 10,000 H100s, along with 10,000 A10s. This is in stark contrast to the 2,048 lower-powered NVIDIA H800 GPU chips that were initially claimed.

Yet, one thing that sets DeepSeek apart from other AI startups is its self-owned data centers, where others rely on external cloud providers. It also underscores the importance of open-source AI models such as Meta’s Llama, which was instrumental in the development of DeepSeek’s R1.

TechDogs-"An Image Depicting The Costs Incurred By DeepSeek According To The Report By SemiAnalysis"
Moving on to other issues with DeepSeek, security researchers from Cisco and the University of Pennsylvania reported that they tested DeepSeek’s R1 model with 50 malicious prompts designed to elicit toxic content—DeepSeek’s model failed to detect or block even a single one.

Essentially, the report noted that DeepSeek R1 exhibited a 100 percent attack success rate, meaning it failed to block a single harmful prompt.

“A hundred percent of the attacks succeeded, which tells you that there’s a trade-off,” said DJ Sampath, the VP of Product, AI Software and Platform at Cisco.

“Yes, it might have been cheaper to build something here, but the investment has perhaps not gone into thinking through what types of safety and security things you need to put inside of the model.”

TechDogs-"An Image Showing Attack Success Rates On Popular LLMs As Observed In The Study By Cisco And The University Of Pennsylvania"
These sentiments were shared by other researchers too, including AI security company Adversa AI.

Despite its failed safety guardrails test, DeepSeek maintains censorship on sensitive topics related to China. Contrary to rumors flying around, this censorship remains in place even if the model is run locally (downloaded to a user’s computer).

Selective censorship still is the tip of the iceberg when it comes to issues related to DeepSeek, as governments around the world are a little iffy about using the tool, owing to data being shared with Chinese servers.

As such, Italy has banned DeepSeek, Taiwan’s Ministry of Digital Affairs warned government agencies from using it, South Korea’s and Netherland’s information privacy watchdog plans to enquire how the company collects and manages personal information, and numerous U.S. government agencies have restricted the use of the model for official purposes.

This included the entire state of Texas, the U.S. Navy, the Pentagon, and others.

Other than data privacy and espionage fears, Japan’s METI (Ministry of Economy, Trade and Industry) voiced its concerns about the increasing number of GenAI models adding strain to power grids and the difficulty in predicting how demand will change with newer and more powerful models coming up. This also adds concerns for achieving zero carbon emission goals.

All this comes in addition to previous developments, where Microsoft and OpenAI had accused DeepSeek of plagiarizing portions of OpenAI’s code and its proprietary models to train its system through distillation.

At the same time, researchers from UC Berkeley, which is also a contributor to the newly launched Oumi—the world’s first unconditionally complete AI platform—said they replicated DeepSeek-R1-Zero’s reinforcement learning capabilities for just $30.

Do you think DeepSeek will be able to challenge OpenAI for the top spot in the GenAI sector?

Let us know in the comments below!

First published on Tue, Feb 4, 2025

Enjoyed what you've read so far? Great news - there's more to explore!

Stay up to date with the latest news, a vast collection of tech articles including introductory guides, product reviews, trends and more, thought-provoking interviews, hottest AI blogs and entertaining tech memes.

Plus, get access to branded insights such as informative white papers, intriguing case studies, in-depth reports, enlightening videos and exciting events and webinars from industry-leading global brands.

Dive into TechDogs' treasure trove today and Know Your World of technology!

Disclaimer - Reference to any specific product, software or entity does not constitute an endorsement or recommendation by TechDogs nor should any data or content published be relied upon. The views expressed by TechDogs' members and guests are their own and their appearance on our site does not imply an endorsement of them or any entity they represent. Views and opinions expressed by TechDogs' Authors are those of the Authors and do not necessarily reflect the view of TechDogs or any of its officials. While we aim to provide valuable and helpful information, some content on TechDogs' site may not have been thoroughly reviewed for every detail or aspect. We encourage users to verify any information independently where necessary.

Join The Discussion

- Promoted By TechDogs -

IDC MarketScape: Worldwide Modern Endpoint Security for Midsize Businesses 2024 Vendor Assessment