Wow this has spread fast in the last couple days. Its been a development since around Christmas when they released v3 but slowly exploded after their reasoning model on inauguration day.

The funniest thing about this is the fact Deepseek is open-source and their entire team publishes full research papers spilling the secrets and formulas they rely on. A Chinese lab is leading democratization of AI

And now Bytedance and others are popping up with similar LLMs.

Years of narratives on Chinese AI being X years behind and researchers lacking innovation destroyed in a week. They're basically doing this handicapped because of chip sanctions too

that's kinda fire tbh.

This has even hit PhD holders and professors who were at prestige universities and are leaving because of geopolitics/discrimination. Its been very common now to see a new story pop up of a veteran scientist with decades in the U.S going home to China to take up a cozy department head at their university.

3 days ago: scmp.com/news/china/science/article/3295774/chen-jing-award-winning-computer-scientist-and-blockchain-expert-returns-china-us

3 weeks ago: scmp.com/news/china/science/article/3293288/ex-dassault-aerospace-physicist-leaves-us-china-lead-new-energy-research

Those are just two quick examples. I can share years worth of this

how much of the public internet has openai used to train their models? it’s fair game atp

venturebeat.com/ai/why-everyone-in-ai-is-freaking-out-about-deepseek

Catch up? They’ve surpassed us and will continue to do so

And this isn’t infrastructure or something that takes decades and trillions to tear down and build up. This is just s*** behind a computer screen

Americans too busy crying about work from home because they desperately want to get keep getting paid $100/hr on top of RSU’s to play video games, make tik toks, and eat out.

What’s wrong with wanting to work from home

Okay boomer.

it's not open source the training code and the data aren't available only the research paper. to be fair the paper shares a lot more than most papers but still

Open source will always catch up and win

it’s enough for people to replicate it isn’t it? that’s all that matters

arxiv.org/pdf/2501.12948

people on twitter are all running it themselves. you just have to pay for API tokens, which are extremely cheap tbf.

five big booms for china

"After thousands of RL steps, DeepSeek-R1-Zero exhibits super performance
on reasoning benchmarks. For instance, the pass@​1 score on AIME 2024 increases from 15.6% to
71.0%, and with majority voting, the score further improves to 86.7%, matching the performance
of OpenAI-o1-0912." I'm listening

It’s ok guys, we’re just gonna throw another $600 billion at AI and hire a bunch of H1Bs.

Well surely catch up.

Did this nigga @op really delete my post?

very interesting !

Niggas got me f***ed up on gooooooooood f*** china till the death of meeeeeeeee

delete this, p****

yeah i have some versions of it running on my mac (google LM Studio it’s super easy to set up and free)

but i mean easyfor researchers to reproduce the same work/make use of the same techniques etc so that others can build on their work, since it goes into detail about how it works and how they did it, which is how scientific research is supposed to work (afaik?)

compare that paper with openai’s gpt4 paper which barely gives any actual info about how they created it at all despite being 100+ pages (mostly self congratulatory “look how good this s*** we did is”) — arxiv.org/pdf/2303.12712

BP poster killin me

anti woke police in china is crazy

We love Panther!

Chinas been on the forefront of computer vision for the last ten years, they realize that the technology itself isn’t what’s important, it’s the application of it. If you really want to be the leader in the ai space you need to make it easy for anyone to start playing around with it.