- cross-posted to:
- technology@lemmy.zip
You must log in or register to comment.
the silly thing is making AI cheaper will pop the western tech bubble even though it’s good for productivity
China’s DeepSeek Is Beating Out OpenAI and Google in Africa
By making AI cheaper and less power-hungry, DeepSeek has put the technology within reach of millions of people.
bUt At wHaT cOsT
Unironically all the issues associated with the technology in general. Just open-sourcing the models doesn’t solve all the problems inherent with them, and as @Wheaties@hexbear.net noted in her pull quotes:
The vast majority of data in Africa hasn’t been digitized, so contractors across the continent are paid to gather agricultural, medical and financial records, as well as audio in Yoruba, Hausa and Nigerian-accented English
Like, is it better that OpenAI or Google doing it? Sure, but that doesn’t necessarily make it good.
Unironically all the issues associated with the technology in general.
Oh yeah, definitely, those would be a legitimate complaint. As usual in tech, solve one problem and two more appear… unfortunately making it less power-hungry can have the side effect of increasing its total power consumption
Researchers have also written about the ethics. Is deploying these sorts of tools to resource-scarce environments going to actually help or will it simply justify decreasing their resources further? Why invest in training teachers, medical workers, and agricultural extension agents if you can just get a chatbot to teach lessons, give medical diagnoses, and tell you when to plant your crops? Are we just going to undermine social relations further by replacing work that plays intangible roles in maintaining communities with software that can only perform the strict job description? And that assumes that it can perform. What happens if your AI extension agent gives you bad crop advice and you lose your harvest?
I’m still convinced that LLMs are mostly a solution looking for a problem and their appeal would be much lower if our ruling ideas weren’t dogmatically aligned with automating away all workers.
It seems that EqualyzAI, the main company mentioned in the article, isn’t even using a hosted version of DeepSeek. They’re running the model themselves on their own hardware, so nobody but EqualyzAI is seeing this data, not even DeepSeek or Huawei.
(Edit: Also EqualyzAI only has to pay for the server costs of running the model, they don’t have to pay DeepSeek anything to be able to self host it.)
Who’s doing what is kind of fuzzy - as worded, it implies that Huawei is doing the digitization and ingestion and EqualyzAI is getting a model produced with custom training weights. I would assume that whatever’s getting integrated is winding up as part of DeepSeek sensu lato because why wouldn’t it?
Even if there weren’t potential issues with that, the broader problem is the uses these models are being put to - see my other comment below. It’s one thing to use an LLM for sentiment analysis or coding help and quite another to use them for actual advice or information that might significantly impact decision making.
Making AI cheaper and less power hungry stands to put the world’s most in-demand technology within reach of millions more people — and to empower African startups to design products for African users.
Uh, I think the bottle neck is less design and more production? Unless by “product” they mean computer code…
[chief solutions architect for Huawei Cloud in sub-Saharan Africa, Harrison Li] clicked through a slide deck, presenting packages tailored to all levels of users and businesses: a free tier, pay-as-you-go hourly rates for DeepSeek models hosted on Huawei Cloud and more compute-intensive options for developers building chatbots and apps [emphasis mine]. For governments, he explained how private cloud systems could be physically installed in offices and ministries.
…they mean computer code. Note there isn’t an example of what these governments can actually use the programme for.
To some critics, this carries ominous echoes of Belt and Road programs that helped some poor countries build critical infrastructure like ports, highways and airports, but left them heavily indebted and financially dependent on Chinese suppliers.
As opposed to before when their suppliers were…? If these countries had funded their infrastructure projects through the IMF, pretty sure they’d still be buying Chinese goods.
that’s just nitpicking, here’s the interesting bit:
For African startups like EqualyzAI, DeepSeek is “orders of magnitude” cheaper than competitors, Adekanmbi said. DeepSeek Chat, for instance, charges 27 cents to process one million tokens of query sent and $1.10 for every million tokens it generates in response. OpenAI’s GPT-4o charges $5 to process the same amount of tokens of query sent and $15 to produce the same amount of tokens in response. If EqualyzAI used GPT-4o, the startup would pay about $12,500 a month to train a small-language model for an e-learning platform, as opposed to the roughly $2,700 per month it now pays DeepSeek for the same task.
So even charging x15 times as much, OpenAI is still running at a loss. A big one. But… even considering that DeepSeek is a more lightweight/efficient programme and China overall is rapidly expanding their electricity output… it still seems hard to imagine any profit is actually happening. About the only actual benefit I see here is this:
EqualyzAI’s engineers used DeepSeek’s open-source architecture as scaffolding to start creating specialized small models as well as automated smart assistants. The vast majority of data in Africa hasn’t been digitized, so contractors across the continent are paid to gather agricultural, medical and financial records, as well as audio in Yoruba, Hausa and Nigerian-accented English. EqualyzAI then trained individual models on the relevant datasets, and tweaked the open weights — the code that instructs an AI model to emphasize or ignore certain information — for each customer. The resulting chatbots and apps are now being used by fintech companies, e-learning platforms and health-care startups. Like all companies that build on DeepSeek, they can choose to either host their products locally and pay for computing and storage infrastructure, or go through providers like Huawei. EqualyzAI does the former.
It keeps the big US firms from building up models in other languages, keeps the servers (relatively) local and in the hands of people who at least actually live alongside the populations who are going to be impacted by it.
left them heavily indebted and financially dependent on Chinese suppliers.
Is there even a single instance where this is true because I heard about the Chinese just straight up forgiving numerous debts
yeah thats usually just propaganda. also open models can be run on infrastructure anywhere so idk what they’re talking about
Nope. In every instance where a debt could not be repaid China either restructured it or outright cancelled the debt.
The reason for this isn’t altruism, it’s rational self-interest. China wants to keep doing productive business with these countries and a bankrupt country shackled with unpayable debt makes for a bad business partner, they will have nothing to offer and little to no purchasing power. It is in China’s interest to help the global south continue to develop economically.
If that means forgiving their debts then that is a small price to pay for a long term, lucrative, mutually beneficial trade relationship.
The reason why the West doesn’t think like this is because the West doesn’t have economies based on real production and development, it has vampire economies based on finance, rent and extraction of super-profits. They squeeze as much as they can out of a country until there is nothing left to squeeze and then blame those same countries for their own poverty on account of not being civilized or capitalist enough, exactly like how when a bigger corporation buys out a smaller firm then runs that business into the ground and sells off its assets.
Also, the West is run by bankers and the entire Western culture has been infected for centuries with the mentality of the banker in which debts are practically worshipped and seen as sacred and the idea of forgiving a debt is an unthinkable sacrilege, a moral peril.
You can also avoid paying trillions to Nvidia is you use Deepseek.The CEO was basically begging Trump to let them sell chips in China while they still can.
From your second quote:
Like all companies that build on DeepSeek, they can choose to either host their products locally and pay for computing and storage infrastructure, or go through providers like Huawei. EqualyzAI does the former.
So that means that DeepSeek is not getting a cent from this company. It’s open-weight, meaning if one has sufficiently powerful hardware they can just run DeepSeek, unlike OpenAI state of the art models, which can only be run by companies that contract with OpenAI to get the weights (as far as I know, this is basically just Google (Vertex) and Amazon (Bedrock)).
But… even considering that DeepSeek is a more lightweight/efficient programme and China overall is rapidly expanding their electricity output… it still seems hard to imagine any profit is actually happening
I think DeepSeek is absolutely burning money. Right now, almost all Chinese models are all open-weight. I’ve seen numerous hypotheses for why this is the case, but I think the one that convinces me the most, at least for DeepSeek, is that they’re doing it as advertising/recruiting. But the revenue that DeepSeek has is only from charging per token on their API as described in your first quote, and they’re competing with every other GPU provider for these prices, so it’s an aggressive race to the bottom. It’s possible that DeepSeek is even running this at a loss to get more training data from people using their API.
In any case, DeepSeek has made a lot of innovations relating to doing more training with less power, because they are currently relatively GPU-poor. NVIDIA chips are hard to come by in China and so DeepSeek can’t really buy any more of the top tier models than they already have. Some of these are used for running the inference for the API, and some are used for the training. But even with all of these optimizations, it costs a lot of money to train an LLM, and it’s hard to imagine that with how often they’re releasing models, they’re actually breaking even, given that at best they have small margins on their API.
