Looks like it's happening. A Chinese company released an "AI" that works just about as well as OpenAI, but it's actually open-source and runs on dramatically less-powerful hardware (like, "individual PC" as opposed to "acres of servers").
Quick correction: you still need server-level hardware to run DeepSeek, but it's orders of magnitude less expense. I have *no idea* what the knock-on effects will be, whether or not this'll ultimately reduce or increase AI slop and power wastage.
Second correction: it's not really open source, it's open model. The code is proprietary, but the model can be trained on data of your choice once you have the software
Yeah, I'm starting to look at everything now and WOW. I'm not super informed on markets but even to someone like me this looks... well, bad if you're into AI, good if you're everyone else I guess?
Every time I see those commercials with people having conversations with their phones I wonder who these people even are? Just google whatever you wanted to know.
The thing about Deepseek is that it's (apparently) open source, and also incredibly cheap to run. So of course it's going to be disruptive to OpenAI's effective monopoly.
Well it’s fun to see OpenAI slip but this suggests that AI as a concept is much *more* durable and potentially disruptive of human jobs than we thought, sadly
I don't think that's necessarily the case. It means there's no moat for the big existing LLMs and thus no more reason to artificially hype up their supposedly disruptive potential.
But if anyone can run a free LLM from their home, a lot the arguments about compute power, privacy, and energy use being prohibitive seen less convincing. Every firm will get one
What I think is likely to result is that we'll stop hearing about what a labor panacea AI supposedly is and it'll normalize to something more aligned with its actual capabilities, which are potentially interesting for certain niches
The only thing I'm actually excited about using generative AI for is sending out bullshit emails. It'll revolutionize office productivity by automating politeness /hj
Plus, you can apparently run a copy of the model locally with not much more than typical consumer hardware? I googled and saw several blogs giving tutorials to install Deepseek models on your PC.
If this works out (though I haven't had a chance to try it yet), then OpenAI's entire business model of holding central control of the model and making money by charging people to use the API falls apart. It's like if Windows tried to charge a subscription in a world where Linux exists.
Every one of them trying to inflate the bubble at any cost, on whatever flimsy pretext they could think of in the moment. They all completely deserve this crash, it was always inevitable
Hilarious. Oh, US tech oligarchs, did you spend the last 3 years purging minorities and the older, experienced workers who built your tech stacks - creating massive shortages of talent and knowledge and distracting everyone into yes-man BS behavior and CYA - while China didn’t? sucks to be u lol
Ok at first I thought you were talking about me and couldn’t have agreed more but then realized it was about AI and still I can’t agree more. I skimmed an article and I am full of a sadistic joy. This is quite the role reversal happening that literally anyone could have seen that isn’t a tech bro
Comments
Both?
https://aleksandarhaber.com/install-and-run-locally-deepseek-r1-ai-model-on-windows/
They're already charging a subscription fee for using the MS Office set of programs. While Open Office and Office Libre exist.
https://www.dazeddigital.com/science-tech/article/55252/1/cryptobros-3-million-dune-book-believing-copyright-nft-jodorowsky-spice-dao?amp=1