Deepseek
2025-01-25 17:36:55.629587+01 by
Dan Lyke
9 comments
Adrianna Tan summarizes the "Deepseek" news:
@kims short version: an eccentric Chinese billionaire set out to try to make something better than ChatGPT, apparently accomplished it with way fewer resources than the U.S. companies are throwing at things, made some breakthroughs, published a paper revealing all the secrets, and showed that you can do what they do for a fraction of the price and resources, which is going to mean that the U.S. companies are going to struggle to raise as much money or will be expected to do much more with less, which means an AI winter might be coming, or that we really can’t compete as well as we think. But also frankly a lot of weird ‘I can’t believe the Chinese can actually do anything’ in this narrative.
This is especially bad for the nerds because their recent wins (cash grab and power) are pegged to them getting to ‘AGI’ and ‘world dominance’ first. I think they didn’t think there was any real competition. This is sort of a Soviets getting to space first type of moment
[ related topics:
Space & Astronomy Current Events Currency Artificial Intelligence
]
comments in descending chronological order (reverse):
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
For whom the ring tones, pointing out that, much like ring tones, when the barrier to them is artificial there's lots of money to be made, but when China is casually tossing out free models and anyone can run them on their own hardware, the value shifts to actual solutions.
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
RT Brandon Rohrer @brohrer@recsys.social
turns out it really was a conspiracy by Big Model to sell more parameters
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
404 Media: OpenAI Furious DeepSeek Might Have Stolen All the Data OpenAI Stole From Us
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
Where's Your Ed At: Deep Impact, talking a lot about the incentives which have kept the US players from trying to be efficient, and some mea culpas on earlier statements. Good read.
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
RT Baldur Bjarnason @baldur@toot.cafe
I seem to be the only one who remembers that LLM output is generally not covered by copyright protection so whatever it is that DeepSeek has been accused of doing absolutely isn't a form of copyright or intellectual property infringement. Terms of Service violation at the most. 🤷♂️
#Comment Re: Deepseek made: 2025-01-29 00:47:29.436802+01 by:
Dan Lyke
RT Justin Pot @jhpot@mastodon.social
Oh no, OpenAI! Are people using your intellectual property against your will? That must be super fucking hard for you…
RT Adam Codega @adamcodega@irrelephant.co
How ironic if Deepseek stole intel from OpenAI for their development.
Maybe OpenAI should have had a robots.txt file in place
RT lcamtuf @lcamtuf@infosec.exchange
OpenAI: how dare others take our... um... copyrighted stuff without asking
to make their own LLM
RT Levi Kornelsen @LeviKornelsen@dice.camp
@kajer @nixCraft - While it's possible they're inflating hard, they've shared how they did this, and it's entirely reasonable, and a big part of it adds up to "Nobody else was doing truly serious (like, at the machine code level) optimizing work AT ALL, they were just throwing money down a hole labelled Going Bigger".
Quoted in Ars Technica’s article on tarpits for AI crawlers:
I am quoted at length in a new article on Ars Technica (a website I have been reading for years and years) on an in-depth article on tarpits for AI crawlers. But people who know my MO will already guess that even being quoted at length is only the tip of the iceberg and Ashley Belanger generously agreed for my to publish my full answers and her questions here.
Run DeepSeek R1 privately on your computer (YouTube video), Via.
(Lots of stuff out there on running DeepSeek on a Raspberry Pi, so much Google-able that I haven't tried to narrow it down.)
#Comment Re: Deepseek made: 2025-01-28 00:09:08.563721+01 by:
Dan Lyke
RT Laura Manach :bongoCat: @cmconseils@mastodon.social
i cant believe ChatGPT lost its job to AI
Karen Hao's Bluesky thread on lessons to be drawn from Deepseek
#Comment Re: Deepseek made: 2025-01-27 19:16:15.027359+01 by:
Dan Lyke
RT Peter @peter@thepit.social
it's funny, if generative AI were actually useful, news of a cheaper, more accessible model would be hugely **positive** for markets! oh, this technology we all thought was prohibitively expensive is actually super affordable?? excellent, let a million products bloom!
#Comment Re: Deepseek made: 2025-01-27 19:16:15.027359+01 by:
Dan Lyke
RT Scalzi @scalzi@mastodon.social
Oh no the Chinese are doing the useless thing that gives wrong answers cheaper
CNN:
A shocking Chinese AI advancement called DeepSeek is sending US stocks plunging
Pivot To AI: DeepSeek R1 is wildly overhyped, but you can try it at home