Explorer

Have Yinz Been Following the DeepSeek News?

Feb 5, 2025

So, basically out of nowhere, a relatively unknown Chinese AI company called DeepSeek showed up and shook things up in a way Iwe haven’t seen since ChatGPT first went viral. Something something may you live in interesting times?

TL;DR Takeaways

DeepSeek Dropped a Surprisingly Capable Open-Source LLM Called DeepSeek-R1 Under a Permissive MIT License.
It’s Cheap and Powerful, Claiming Comparable Results to Top Models but Trained at a Fraction of the Cost.
Their App Skyrocketed to #1 on Both the App Store and Google Play Store in the U.S.
Red Flags Everywhere — Censorship, Security Concerns, Possible Shady Training Practices.
Microsoft Is Both Investigating and Hosting Them, Because of Course They Are.

The Launch Heard ‘Round the World

DeepSeek released their model, DeepSeek-R1, onto Hugging Face with barely any warning. Unlike most “open” models, they slapped it with an MIT license, meaning anyone could use it commercially, no strings attached. Even wilder? It isn’t just open, it is (allegedly) good. DeepSeek claims it performs well on tough benchmarks like MATH-500 and SWE-bench Verified, which puts it in the same conversation as OpenAI’s models. And they say they trained it for just $5.6 million, using older Nvidia GPUs. For comparison, OpenAI and Anthropic reportedly spend hundreds of millions on their top-tier models.

Naturally, the Internet Had Hot Takes

Marc Andreessen called DeepSeek “one of the most amazing and impressive breakthroughs I’ve ever seen.”
Yann LeCun from Meta jumped in to argue that this proved why open-source AI was the future.
Garry Tan took a more measured stance, saying cheaper models would drive infrastructure demand, helping cloud providers and hardware companies.

Top of the Charts

Meanwhile, over on Hugging Face, things exploded: more than 500 derivative models popped up based on R1, and downloads hit 2.5 million+ in just a few days. DeepSeek’s own AI assistant app also skyrocketed to #1 on both the App Store and Google Play Store in the U.S., which only fueled the hype even more. I wasn’t one of the people to download it, something about it just made me think it wasn’t something I wanted monitoring my phone activity, so I can’t speak to it’s effectiveness personally, but other user reports seem positive?

The Red Flags (Because There Are Always Red Flags)

Of course, it wasn’t all sunshine and app store charts.

Researchers quickly noticed that R1 heavily censored politically sensitive topics. Anything related to Tiananmen Square, Taiwan, etc., got deflected with awkward nationalistic replies (I’m shocked, shocked I tell you, well not that shocked). Security concerns piled up fast too. Hundreds of U.S.-based organizations, especially ones with government contracts, blocked DeepSeek’s services entirely, citing data risks.

And then things got even messier. According to The Information, Microsoft launched an internal investigation into whether DeepSeek illegally trained on OpenAI API outputs — a big no-no under OpenAI’s terms. At the same time, Microsoft added DeepSeek-R1 to their Azure AI Foundry marketplace, saying it “passed rigorous safety evaluations.”

Capitalism, baby.

Oh Yeah, They Also Dropped Image Models

Because why not, right? Toward the end of the month, DeepSeek released Janus-Pro, a family of image generation models they claim outperform DALL·E 3. These were also dropped on Hugging Face, ranging from 1B to 7B parameters. So expect even more AI-generated junk clogging your Instagram feed soon!

Final Thoughts

I think it’s still too early to tell for sure, but I am willing to bet that once the hype wears off DeepSeek isn’t going to live up to it’s claims. But even if it doesn’t, the fact that a relatively tiny team with limited resources can create something that gets this much attention, this quickly, says a lot about where the AI world is heading. Fast, chaotic, and global.

Grab your popcorn.