wbftw
𝐿𝒶𝓃𝒶 "not yet begun to fight"For anyone wondering why the Nazis on Twitter, Meta, and the US State are having a meltdown today it's because China dropped about a half dozen brand new AI models that outperform anything ChatGPT/Grok/Claude/Llama can do, take a fraction of the time and money to create and upkeep, and made them all open source, ruining the AI market.
You Nazi fucks wanted a trade war. You got one. #FAFO
☮️ & 🍺 on 🌏
Dave Spector@Lana they’re not open source they’re just mostly freely usable.
Meta has tricked everyone into believing that since they let people use Llama for free it’s “open source.” Others have adopted that nomenclature.
Open source means “comes with all the source materials.”
To date, you can count the number of orgs that have released all the training-data/code/weights/tools and so on that you’d need to recreate any large LLM from scratch on one half of one hand.
@Lana now, having said that, if they really did create and train this LLM for less that $6 MM — and it’s really quite impressive, the 70billion param model runs an M4 MacBook pro without breaking a sweat — it really puts the lie to Altman, Musk, and the rest in terms of their expenses and how much money they’ve blown and investors should be suing the crap out of all of them.
@Zitron predicted this almost a year ago.
Misuse Case@Dhmspector @Lana @Zitron I think part of what’s been going on with American AI companies is a built-in kickback to chip makers like Nvidia
DrYak@Dhmspector the expensive data centre were mostly a form of gatekeeping against upcoming competitors.
@dryak if DeepSeek have figure out some clever optimization all that power and build out was a lot of cash they incinerated.
@Dhmspector It's a bubble.
"Incinerating lots of investors' cash solely for the personal gain of a couple of tech bro billionaires" kind of comes with the territory.
"Incinerating lots of investors' cash solely for the personal gain of a couple of tech bro billionaires" kind of comes with the territory.
@dryak true… but this is a lot more than usual and since Web2.0 LPs are a lot less happy with losing this kind of cash.
@Dhmspector no, it's open source. You can literally modify its algorithm.
David PThe training data/test data etc is the only thing unavailable.
crypticcelery@pewnack @Lana @Dhmspector that is precisely the point. Models like DeepSeek-R1 are just “Open Weight”, the training data and exact methods remain a secret. Also, there are some weird licensing quirks with models like LLaMa, not sure about this one (no expert).
The point is: It cannot be replicated, and we have no idea about the ethics (copyright, labour, censorship, bias, …) of it = not really open.
Though it is still fun to see the genAI bros freak out.
@Lana this is a great explainer on why just having open weights isn't the openness that people might think it is.
You and I are not going to get a lot of out having access to these weights.
You Ask, I Answer: Open Weights, Open Source, and Custom GPT Models? - Christopher S. Penn - Marketing AI Keynote Speaker
You Ask, I Answer: Open Weights, Open Source, and Custom GPT Models? In today's episode, Joseph asks if it's possible to create your own custom GPT model
GaMe@Dhmspector that's a good point. Are there any truly open source LLM? @Lana
Jonathan Yu@Dhmspector @Lana Yes, you're correct, but the folks at Huggingface are working on making it reproducible
Lorry@jawnsy @Dhmspector @Lana humph, #HuggingFace turned me down without interview for a role I was very qualified for, I am sure it was because #ageism - So I'm not happy with them.
It doesn't justify the money burned but some costs in Western public models are in post-training #TrustAndSafety work. Constantly having humans check and tweak output, running adversarial tests, and simply #training the #AI in specific skills. I guess that won't come with the #Chinese ones.
I live in hope though!
@Dhmspector @jawnsy @Lana yeah #ageism is systemic. Getting a job in tech if you are over 50 feels like being a kid's party magician with leporacy.
I have been advised to cut 20 years off my resume but that's by people who are in their early 30s, and it just doesn't really work.
Last time I applied to #Mozilla for a role I would probably have been the best candidate available in the world for, I was rejected without even a pre-screen recruiter interview. It's all very silly really!
@Dhmspector @jawnsy @Lana I for one welcome our #AIOverlords with open arms 😁
I just hope that they introduce Universal Basic Income.
Sadly, #UBI is a fantasy that the billionaire brats club spews to try to beguile those susceptible to magical thinking to make them think the destruction of jobs has an upside.
There will never, ever be UBI in this country — the puritanical/calvinist foundations of the place would never allow it.
In the American work-ethic canon, if you can’t afford to live its “God’s will” and you should just die.
Fedo ¶@Lana you misspelled “The Very Very Bad China”
Chill
Gregory
Paul Sochacki@Lana best news I've heard all week- thank you for sharing!
Rob
AlgoCompSynth by znmeb 🇺🇦@Lana Why does this story remind me of "Colossus: The Forbin Project"?
Miles Goodhew
They did Nazi that coming ;-)
Eniko (moved ➡ gamedev.place)@Lana this is very funny but I was hoping AI would collapse and go away and I don't think this will do that 😔
Aubrey JonesThere's really no technologically advanced future without AI, including generative AI and LLM. Everybody who's ever wanted a voice interface to a computer (e.g. Star Trek) needs LLMs.
However, in a socialist society where the right to exist is not tied to productivity, divorced from a profit motive to create markets, AI is no economic threat.
Even artists and writers (like me) wouldn't care what the machines derive if we could just afford to eat and work on our own projects.
Klarname 🌈@Lana it says "semi open source" and dont specifiy. Chinese Big Tech companies dont habe a track record of big open source repos.
Firnin 
R.L. LE@Lana I'll start getting interested, when they need only a fraction of the electricity, too? 🤔
Sepideh@herrLorenz @Lana Does that part about less graphics GPUs mean that it is taking less electricity? I assumed it did.
Magess 
@herrLorenz @Lana well they trained it with 1/5 the number of GPUs, so that is a lot less electricity
bazzete@herrLorenz @Lana for training yes. And also running it. The smaller distilled versions of it can run locally on a nowadays average pc and maybe even top shelf smartphone. So hopefully this trend continues and these models will be run locally in your smartphone and will be good enough for all your general tasks.
Ra@Lana you could theorize that this is a crushing real world response to TFG's performatve order to the CIA to release their less than confident 'lab leak' investigation.
aproitz
Andrew@Lana @tayfonay i tried it out Friday for some work stuff and yep - absolutely on par with the latest Chatgpt for my particular use case. Meanwhile, Altman and Musk are bickering about the half trillion dollars supposedly needed to make a better model in the US. 🤦🏼♂️ Such a waste of time and resources.
Bruno Nicoletti
violetmaze@Lana What's the environmental cost? I haven't used AI as all I hear is how much water it uses and how it's so energy intensive.
cultdevMagess :heart_ace: (@[email protected])
@[email protected] @[email protected] well they trained it with 1/5 the number of GPUs, so that is a lot less electricity
@violetmaze @Lana If the model didn’t cost as much that means the amount of computational overhead, hardware, and power was that much lower.
Winfried is Anti Fascism 🏳️🌈@Lana
I also love the timing...
I also love the timing...
Proto Himbo European@Lana It's not like we didn't know this was coming. Even I, a non-AI-professional, have seen a couple of articles in the past six months about open source AI models that match the performance of chatGPT, Gemini, etc. at a fraction of the energy cost.
I think a lot of CEOs put all their eggs in this basket when the entire world was telling them not to put all their eggs in this basket.
Phil Thane ✅@Lana
Meanwhile Brexit island Starmer is talking nonsense about us becoming an AI powerhouse.
@Maker_of_Things
Meanwhile Brexit island Starmer is talking nonsense about us becoming an AI powerhouse.
@Maker_of_Things
Kevin Karhan 
Pep@Lana Generative AI may be bad, but I'm always for open source undercutting paid products from large companies.
Marcus Marcius Marcellus@Lana Srsly? I mean, I hate AI and all, and the Chinese regime is hardly a great benefactor of humanity.
But reducing the value of asshole US billionaire techbros’ assets to nil in one fell swoop is, let’s be honest here, absofeckinlutely legendary.
Juan Per¢ent,🇲🇽 🍉,🇻🇪@marcas @Lana @marcas @Lana" China has contributed close to three-quarters of the global reduction in the number of people living in extreme poverty . At China’s current national poverty line, the number of poor fell by 770 million over the same period. ( in the last 40 years) "
Improving the material conditions 800 million people is a net benefit to humanity perhaps?
Improving the material conditions 800 million people is a net benefit to humanity perhaps?
Shaman
Dmitry Tantsur@Lana @dacig @marcas this aspect is truly impressive, but net benefit for *humanity*? I assure you that Ukrainians and generally Eastern and Central Europeans are less than impressed that China is more or less singlehandedly keeping afloat the aggressor destroying Ukraine and threatening the rest of Europe.
MaineC
Aurora 🐌