Budget AI Model DeepSeek Overtakes ChatGPT on App Store

[Moka Akashiya]

And giving China more free data of yours. Which is how they were able to do this dirt cheap originally 😩

You are giving all data to OpenAI and Apple, that tracks all your open apps and checks them in their server to allow them to open, why do you care about China at this point? You can use local model if you want for free and see full implementation, so this product is already more trustworthy when used this way.

It's also very funny with ignorant people who think that their prompts with questions are valuable data when there is a sea of already organized public data. It seems that all such people worry about is the nationality of the authors, even when you are given a completely free and open product that does not require sending data to a server. Waiting for better arguments if you disagree.

 

[ThisIsMike]

This is actually fantastic news for Apple. This would support the idea that if they put enough investment and manpower behind it, Apple can still catch up to, or even surpass, current AI models, and they can do it while spending less. Fingers crossed they actually execute.

 

[jonnyb098]

You are giving all data to OpenAI and Apple, that tracks all your open apps and checks them in their server to allow them to open, why do you care about China at this point? You can use local model if you want for free and see full implementation, so this product is already more trustworthy when used this way.

If you wanna share your data freely with China then have at it pal!

 

[SgtPepper12]

If you wanna share your data freely with China then have at it pal!

You do realize that all LLMs are trained on mostly the same massive piles of data scraped from the internet with very little regard for copyright, privacy or ethicality, right? Your average spaghetti recipe prompt is of absolutely no value whatsoever to DeepSeek or OpenAI, given the vastness of data that's freely available already. I'd be far more scared of Google, Microsoft and Meta, that actually have your private documents, messages and e-mails at their fingertips, just wondering whether anybody would really notice if they trained a model on that massive, untouched pool of data. Cause, boy, that would make a big difference.

 

[big_papa_johnny]

More like it interferes with techbros ability to scam $500 Billion away from taxpayers.

The $500bn Stargate project will not be funded by taxpayers.

 

[turbineseaplane]

With where this all seems headed I find myself missing the 1980s and 1990s

 

[Razorpit]

I tried asking DeepSeek about the Tiananmen square massacre. It won’t answer.

Excellent test. That's the bellwether for any "AI" coming out of China.

 

[qnssekr]

Governments around the world will be looking very closely at this indeed. $6 million is nothing. A lot of countries out there will be quietly very keen to reduce their dependence on US tech giants.

What makes you think that price is the ACTUAL price? Do you believe everything that comes out of China?

 

[Razorpit]

Chinese gov can't send you to American jail.

But they can if you ever visit China, or another China friendly nation.

It's crazy how our lives are contained on hundreds of databases across multiple companies and traded/sold.

My Dad just passed away on the weekend, and he always said cash was king, he had a bankbook, no credit card, when he got paid he'd take all his money out and hide it around as he didn't want anyone to know how much he had or what he spent. To him it wasn't anyone else's business, and he is right. Seems impossible in this digital age, but I think about the method to his madness a lot.

Sorry for your loss. 🥹

Try asking for vaccine side effects in ChatGpT and include the test report, chargpt will initially lie and then it let iteslf be corrected. No explaination why it can't count in % or /

Had a similar issue last week. I couldn't find an answer to something using the normal search engines so I used ChatGPT. Later when I found the answer to probably be incorrect, I went backed and asked it to please source it's answer. It couldn't.

I went on to ask why did you feel the need to give me information that you had nothing to back it up with. It had no answers other than basically, I did what I did, I'm sorry. How can anyone trust these LLM's to run anything is beyond me.

 

[H2SO4]

Governments around the world will be looking very closely at this indeed. $6 million is nothing. A lot of countries out there will be quietly very keen to reduce their dependence on US tech giants.

With good reason. The US are just a different kind of corruption - not necessarily a better one.

 

[jimbobb24]

So far I have been unable to log on to try it. Maybe their servers cannot handle the load.

At this point LLMs are only seeing small gains and how to make them isn't mysterious. The question is did they make it so cheaply, did they use an open set of weights to get started?

On websites that compare state of the art models deepseek does seem to be doing pretty well. Curious how robust it is in languages and carrying on a conversation in multiple language, stuff like that. But from what I can see they did a solid job and if they did for so cheap it's great news. The price of LLMs might be falling.

 

[jimbobb24]

Try asking for vaccine side effects in ChatGpT and include the test report, chargpt will initially lie and then it let iteslf be corrected. No explaination why it can't count in % or /

How is the fact that DeepSeek cannot comment on Tinneman square surprising given the incredible limitations and curation of allowable data in the US LLMs. Almost all US LLMs will refuse to answer questions the elite find uncomfortable about many topics. Gemini and ChatGPT wouldn't show me publicly available data because they said it was uncomfortable or some such.

We can point our fingers at Chinese censorship but US LLMs have a huge number of topics they cannot talk about because its makes some people feel awkward.

 

[andiwm2003]

So far I have been unable to log on to try it. Maybe their servers cannot handle the load.

At this point LLMs are only seeing small gains and how to make them isn't mysterious. The question is did they make it so cheaply, did they use an open set of weights to get started?

On websites that compare state of the art models deepseek does seem to be doing pretty well. Curious how robust it is in languages and carrying on a conversation in multiple language, stuff like that. But from what I can see they did a solid job and if they did for so cheap it's great news. The price of LLMs might be falling.

The main question here is not if you trust this or that government or this or that company. It’s also not if you feel cheated because they use your data to train their models. That ship has sailed long ago. Sucks but nothing that can be done about that anymore.

The question is if they have made a technological breakthrough or have they just used some efficiencies derived from existing models (let’s say they just improved the weights of Grok).

If it was a breakthrough then the world has changed from an investment perspective, a national security perspective and a social perspective.

 

[Student of Life]

This is a game changer because it’s free and open sourced. This will impact the market, what investor will fund a company, that its competitor give away the same product for free? Even if it’s 90% of chatGPT it will be good enough for most and it’s free. Who would pay 20+ a month when an equivalent product is free? This will dominate like google products because they are free. One can spend billions but when the completion is giving it away for free, your products value just got destroyed.

 

[Realityck]

https://www.cnbc.com/2025/01/27/power-stocks-plunge-as-ai-energy-needs-questioned-due-to-new-china-model.html

Power companies that are most exposed to the tech sector’s data center boom plunged early Monday, as the debut of China’s DeepSeek open source AI laboratory led investors to question how much energy artificial intelligence applications will actually consume.

Constellation Energy and Vistra Corp. tumbled more than 16% in morning trading. GE Vernova slid about 18% while Talen Energy lost more than 15%.

Constellation, Vistra and GE Vernova have led the S&P 500 this year as investors speculated that AI data centers will boost demand for enormous amounts of electricity.

But DeepSeek has developed a model that it claims is cheaper and more efficient than U.S competitors, raising doubts about the vast sums of money the tech sector is pouring in to data centers.

 

[fatespawn]

I tried asking DeepSeek about the Tiananmen square massacre. It won’t answer.

That was the first thing I was going to do but I can't get past the sign up page. Ask it about the Uyghurs and see what it says.

 

[rovex]

Wait for the day Google lens will be allowed to search people faces. If you’re on social media, best to delete your profiles asap.

 

[CapitalIdea]

same problem as tie Tok I would think. China could be using app to collect info. spy. record. like amazon and Alexia.

I’m sure China will be just fine without insights into your query about suggested sides with a chicken on focaccia.

 

[Frantisekj]

same problem as tie Tok I would think. China could be using app to collect info. spy. record. like amazon and Alexia.

... and anyone else. That is why we probably use Apple as we hope it is using data for its own marketing purposes.

 

[msackey]

same problem as tie Tok I would think. China could be using app to collect info. spy. record. like amazon and Alexia.

The thing is, why are we giving so much trust to the Tech Bros who are just as likely to either sell off our data to the "highest bidder" or basically just violate confidentiality and therefore result in national and personal security risks.

In other words, just because Tech Bros are American citizens doesn't mean they are any better.

 

[thefrost]

Wow! Shiver me timbers!

 

[CapitalIdea]

I saw a fascinating report on this on CNBC. They were able to achieve this with older hardware and 1/1000 the budget we are spending on our closed source (read: to be monetized at every turn) variant. Pretty soon our AI endeavors will resemble our healthcare system.

 

[Frantisekj]

Honestly what can China do with your dancing video data?
You would rather have your data given to US companies so they could pass them to your gov, so that they can technically get any info they want or able to quickly take any action on you?

The video data are the last what they want to get from you. But all other, location, places you visit, people you meet, your interests, habbits, political opinion and so on....

 

[urtules]

the US has no data privacy laws at all.

If I want to compromise on my personal information, like email address and the queries, I would rather trust any western private company than Chinese government.

 

[dannys1]

At this point, with this news coming out, realistically who in the public would fund this? The competition is giving away a comparable product for free, and it consumes less resources. As an investor, the basic question is, how could I monetize this when it's being given away for free and opened sourced so the better question becomes why should I invest in this?

Very true - it'll have shaken up OpenAI's plans no end. The biggest problem isn't so much that this performs as well as their own product it's the fact that the Chinese were able to create it with a fraction of a the cost. The investors are going to want to know why it costs seemingly 50x the price to create the models.