I keep seeing posts about DeepSeek like this one (x.com). Is it just another language model, or is there more to it? Some people are saying it could be a huge problem for Nvidia and big tech. Is that just hype, or is there something real going on?
DeepSeek is a Chinese AI startup that just released its R1 model, and it’s got a lot of people talking. The biggest reason? They trained it for around $5.6 million, which is nothing compared to the billions OpenAI, Google, and others are spending.
A lot of AI companies have been saying that making these models cheaper just isn’t possible, and that they need more powerful hardware. But DeepSeek managed to do it by optimizing their training process. Some big companies that have been relying on expensive AI models are now wondering if they should switch.
That said, it’s not all perfect—DeepSeek R1 is slower, and there are still questions about how it was really trained. But the fact that they pulled this off at such a low cost is sending shockwaves through the industry.
@Meade
One thing people aren’t talking about enough is that DeepSeek is mostly open source. The big tech companies in the US have been keeping their models locked away to protect their profits. Now, DeepSeek is showing that you don’t have to do that, and people are paying attention.
@Dallas
Yeah, that’s a huge part of it. The open-source aspect is a major game changer.
Meade said:
@Dallas
Yeah, that’s a huge part of it. The open-source aspect is a major game changer.
It’s also worth noting that it runs slower because they’re using Nvidia’s restricted H800 GPUs instead of the high-end ones that US companies have access to.
@Breck
But are they really using those, or is that just the story they’re telling? There’s a good chance they have access to way more GPUs than they’re admitting.
Gracen said:
@Breck
But are they really using those, or is that just the story they’re telling? There’s a good chance they have access to way more GPUs than they’re admitting.
That’s possible, but without solid proof, we can’t say for sure.
@Breck
People always assume the worst when it comes to China, even when there’s no actual evidence.
@Breck
They probably just repurposed a bunch of GPUs from old crypto mining farms. Makes sense from an economic standpoint.
Meade said:
@Dallas
Yeah, that’s a huge part of it. The open-source aspect is a major game changer.
What’s really happening is that China is finding ways to move ahead while Silicon Valley is still focused on keeping control of everything. OpenAI and others thought they could keep AI locked up, but DeepSeek is proving that’s not the case.
The reason this is such a big deal is because AI has taken over the tech industry. Companies have spent billions trying to stay ahead. Now, a small team from China has built something better and made it open-source. Why would people keep paying for expensive AI services when this is available for free? It’s got a lot of executives scrambling to figure out what’s next.
@Micah
If this was possible, why didn’t all the big companies see it coming?
Sky said:
@Micah
If this was possible, why didn’t all the big companies see it coming?
A lot of times in tech, people just assume that bigger and more expensive is always better. But sometimes, a smaller and more efficient approach works just as well. This happens all the time in computing—somebody figures out a smarter way to do something, and suddenly everything changes.
Another reason this is making waves is the cost. OpenAI’s most advanced model costs about $2.50 to process a million tokens. DeepSeek does the same for $0.14. That kind of price difference changes everything.
One thing to remember—OpenAI paid the price of being the first mover. They spent billions proving AI could work at this level. Now, companies like DeepSeek can come in and refine the process at a much lower cost.
One issue with DeepSeek is how much filtering it does. It refuses to answer questions about certain topics, especially related to China. It’s great that it’s open-source, but if you’re using the official version, you’re dealing with a lot of censorship.
@liz
That’s true for the official model, but since it’s open-source, people can modify it however they want. There are already projects working on an uncensored version.
I keep looking for details about the data DeepSeek was trained on, but I can’t find much. Anyone know more?
Kale said:
I keep looking for details about the data DeepSeek was trained on, but I can’t find much. Anyone know more?
Probably a mix of publicly available data and whatever other sources they could get their hands on.