Welcome to our community
Be apart of something great, join today!
Coding, Data Science, A.I. catch-All | Grok update goes MechaHitler
- Thread starter nycfan
- Start date
- Replies: 258
- Views: 7K
- Off-Topic
DeepSeek's AI Breakthrough Sparks Microsoft, OpenAI Probe Into Possible Data Breach
Microsoft and OpenAI probe potential unauthorized data use by DeepSeek as U.S.-China AI competition intensifies
finance.yahoo.com
Microsoft (NASDAQ:MSFT) and OpenAI are investigating whether a group linked to Chinese AI startup DeepSeek accessed OpenAI's technology without authorization, Bloomberg reported, citing sources familiar with the matter.
Microsoft security researchers detected individuals believed to be associated with DeepSeek extracting large amounts of data via OpenAI's API last fall. Under OpenAI's terms, developers must pay to use its API, which limits how much data they can access.
As OpenAI's largest investor, Microsoft flagged the activity as a potential violation of OpenAI's terms of service or an attempt to bypass restrictions. DeepSeek's latest AI model, R1, has outperformed U.S. rivals at a fraction of the cost, triggering a selloff that wiped out $1 trillion in market value on Monday.
gtyellowjacket
Iconic Member
- Messages
- 2,057
More is coming out confirming this theory. One way they confirmed it was simply by asking the deepseek model and it said it was trained on Openai.
DeepSeek's AI Breakthrough Sparks Microsoft, OpenAI Probe Into Possible Data Breach
Microsoft and OpenAI probe potential unauthorized data use by DeepSeek as U.S.-China AI competition intensifies
Microsoft (NASDAQ:MSFT) and OpenAI are investigating whether a group linked to Chinese AI startup DeepSeek accessed OpenAI's technology without authorization, Bloomberg reported, citing sources familiar with the matter.
Microsoft security researchers detected individuals believed to be associated with DeepSeek extracting large amounts of data via OpenAI's API last fall. Under OpenAI's terms, developers must pay to use its API, which limits how much data they can access.
As OpenAI's largest investor, Microsoft flagged the activity as a potential violation of OpenAI's terms of service or an attempt to bypass restrictions. DeepSeek's latest AI model, R1, has outperformed U.S. rivals at a fraction of the cost, triggering a selloff that wiped out $1 trillion in market value on Monday.
Essentially the way that Deepseek was allegedly able to do it so cheap was to piggyback on the work that openai did. Openai basically trains on the internet. That's a really big place and it needs an awful lot of resources. But deepseek could train its model basically by querying openai and getting a lot of shortcuts.
So if true, what does that mean for the future? After all China has been stealing the intellectual property of everything from Rolex's to DVDs for decades now. The difference is that there are ways to shut down this training now that Openai knows what to look for. So that big sell-off of Nvidia might be overblown, And there very well still may be a lot of demand for all that hardware.
Last edited:
lawtig02
Inconceivable Member
- Messages
- 3,775
All that seems right to me. And I would add this also means that while Sam Altman might have a very high IQ, he’s still a freaking idiot.More is coming out confirming this theory. One way they confirmed it was simply by asking the deepseek model and it said it was trained on Openai.
Essentially the way they were able to do it so cheap was to piggyback on the work that openai did. Openai basically trains on the internet. That's a really big place and it needs an awful lot of resources. But deepseek could train its model basically by querying openai and getting a lot of shortcuts.
So what does that mean for the future? After all China has been stealing the intellectual property of everything from Rolex's to DVDs for decades now. The difference is that there are ways to shut down this training now that Openai knows what to look for. So that big sell-off of Nvidia might be overblown, And there very well still may be a lot of demand for all that hardware.
superrific
Legend of ZZL
- Messages
- 8,747
It doesn't matter all that much. The idea that NVidia is going to retain a stranglehold on GPU chips is not well-founded.
Better would be for OpenAI to introduce countermeasures to corrupt a poachers model.More is coming out confirming this theory. One way they confirmed it was simply by asking the deepseek model and it said it was trained on Openai.
Essentially the way that Deepseek was allegedly able to do it so cheap was to piggyback on the work that openai did. Openai basically trains on the internet. That's a really big place and it needs an awful lot of resources. But deepseek could train its model basically by querying openai and getting a lot of shortcuts.
So if true, what does that mean for the future? After all China has been stealing the intellectual property of everything from Rolex's to DVDs for decades now. The difference is that there are ways to shut down this training now that Openai knows what to look for. So that big sell-off of Nvidia might be overblown, And there very well still may be a lot of demand for all that hardware.
gtyellowjacket
Iconic Member
- Messages
- 2,057
I should mention about Deepseek that before they "borrowed" open ai's model to train their model, they still had some really good stuff. They have smart folks on that team and they provide some really high-end models, especially coding assistants. They may or may not be part of the Chinese government. Their big issue right now is getting the hardware to train their models to keep up with open AI, Google and Meta because of export controls.
I don't know this, but another issue I think they are having is that announcement got them a lot more customers And they can't get the hardware to handle it. I am getting the equivalent of server busy messages all the time now and I was never getting them before.
I don't know this, but another issue I think they are having is that announcement got them a lot more customers And they can't get the hardware to handle it. I am getting the equivalent of server busy messages all the time now and I was never getting them before.
Here's a do-gooder job for a jack of many trades data wrangler at ProPublica
ProPublica
<p style="margin-bottom: 1em;">ProPublica is an independent, nonprofit newsroom that produces investigative journalism in the public interest. With a team of more than 100 dedicated journalists, ProPublica covers a range of topics, focusing on stories with the potential to spur real-world...
boards.greenhouse.io
- Messages
- 22,040
A Deep Peek Into DeepSeek AI’s Talent And Implications For US Innovation
DeepSeek AI’s rapid ascent signals more than a breakthrough in large language model performance—it is a wake-up call for US innovation strategy. This Chinese startup was propelled by a deeply rooted and increasingly self-sufficient domestic talent pipeline.
This study debunks the notion that Deepseek was developed by relative novices in China (used by some to support allegations of stealing American tech) and points to a critical need for the U.S. to focus on human expertise to avoid losing its lead in the AI and tech sectors.
“… The DeepSeek AI case illustrates a critical blind spot in US technology policy: the erosion of America’s human-capital edge.
Export controls and computational investments are necessary but insufficient. The United States cannot outregulate its way to technological dominance.
Competing in this new era requires a renewed focus on developing, attracting, and retaining top talent, both from within the United States and abroad. If DeepSeek AI is an early indicator, the future of technological leadership will be won not just with faster chips or bigger models but with smarter strategies for global talent competition. …”
BlooVooDoo
Iconic Member
- Messages
- 1,703
It’s early, but all signs suggest this is not likely to be a US strength in the second Trump term.
- Messages
- 22,040
Yeah and long term, even assuming that this second term is a political anomaly from which the body politic recovers, reversing the brain drain will take resources, time (probably including legislation to codify protections against the next Trump-type elected official’s abuse of authority against immigrants) and effort.
superrific
Legend of ZZL
- Messages
- 8,747
No, it's the blind spot in Trump's tech policy.
Sane people have been talking about human capital for years. I certainly know I've been focused on it.
- Messages
- 994
It’s not like China is recruiting the global tech elite either. If the assumption is that China can produce enough home-grown talent to out-compete us, that still (currently) depends on Chinese getting US education for undergraduate and grad school. To say nothing of China’s demographic cliff.
If we can keeps our doors open to talent, the best of India, China etc. will continue to come here. You can’t keep em down on the farm once they’ve seen Karl Hungus.
If we can keeps our doors open to talent, the best of India, China etc. will continue to come here. You can’t keep em down on the farm once they’ve seen Karl Hungus.
superrific
Legend of ZZL
- Messages
- 8,747
But China's population is considerably more than the US and Europe COMBINED. And I don't know all the particulars of their education system, but they obviously develop plenty of talent homegrown. There are a lot of Chinese students in the U.S., but if even 10% of US university students are from China (seems like a stretch, but maybe if you include graduate studies in niche fields), that's still not that significant. China's population is 5x as high as the US.
IIRC you've worked in China so you likely know more than I do about this specific example. I just think the numbers advantage for China is a huge factor.
gtyellowjacket
Iconic Member
- Messages
- 2,057
Deepseek is good, but it has some issues that may keep it from leading.A Deep Peek Into DeepSeek AI’s Talent And Implications For US Innovation
DeepSeek AI’s rapid ascent signals more than a breakthrough in large language model performance—it is a wake-up call for US innovation strategy. This Chinese startup was propelled by a deeply rooted and increasingly self-sufficient domestic talent pipeline.www.hoover.org
This study debunks the notion that Deepseek was developed by relative novices in China (used by some to support allegations of stealing American tech) and points to a critical need for the U.S. to focus on human expertise to avoid losing its lead in the AI and tech sectors.
“… The DeepSeek AI case illustrates a critical blind spot in US technology policy: the erosion of America’s human-capital edge.
Export controls and computational investments are necessary but insufficient. The United States cannot outregulate its way to technological dominance.
Competing in this new era requires a renewed focus on developing, attracting, and retaining top talent, both from within the United States and abroad. If DeepSeek AI is an early indicator, the future of technological leadership will be won not just with faster chips or bigger models but with smarter strategies for global talent competition. …”
-Deepseek does not necessarily have access to the same hardware resources from Nvidia and now Google. There is a reason Nvidia is making all the money in AI right now and its because they have something that others have had a hard time duplicating. There are significant controls in place that make it very hard for deepseek to get those same computing resources. Deepseek acquired a lot of their chips before the export controls were put in place but it will be harder to get new ones.
-Being the best is really hard. There is only one and its probably ChatGPT. Even Google and Microsoft, with all their resources, haven't caught up to ChatGPT. Deepseek probably trained its most recent model with the help of ChatGPT giving them a big shortcut. ChatGPT, and presumably all the other AI companies, put measures in place to prevent that going forward. Deepseek has not come out with a new release since then but that could change tomorrow. All the other big companies have.
-Deepseek has a bit of a nebulous origin. It's supposedly funded by a Chinese venture Capital company that no one has ever heard of. It was literally founded less than 2 years ago. So that means the billions of dollars it takes to train one of these models is coming from some unknown source. They claim that they can train for much cheaper than the other guys but that goes back to maybe sponging off ChatGPT. What that likely means is it's funded by the Chinese government. Are people in the west really going to trust that?
I'm not saying China's talent pipeline isn't robust because it is, but I think it's still a long way behind the United states. They are catching up though in a way the Russians or Europeans or even the Indians never did.
- Messages
- 5,854
This is why we need to eliminate books in school and the department of education, display the 10 commandments on every wall and start teaching all subjects from the Bible.A Deep Peek Into DeepSeek AI’s Talent And Implications For US Innovation
DeepSeek AI’s rapid ascent signals more than a breakthrough in large language model performance—it is a wake-up call for US innovation strategy. This Chinese startup was propelled by a deeply rooted and increasingly self-sufficient domestic talent pipeline.
This study debunks the notion that Deepseek was developed by relative novices in China (used by some to support allegations of stealing American tech) and points to a critical need for the U.S. to focus on human expertise to avoid losing its lead in the AI and tech sectors.
“… The DeepSeek AI case illustrates a critical blind spot in US technology policy: the erosion of America’s human-capital edge.
Export controls and computational investments are necessary but insufficient. The United States cannot outregulate its way to technological dominance.
Competing in this new era requires a renewed focus on developing, attracting, and retaining top talent, both from within the United States and abroad. If DeepSeek AI is an early indicator, the future of technological leadership will be won not just with faster chips or bigger models but with smarter strategies for global talent competition. …”
- Messages
- 994
There are more than 2000 Chinese students at Purdue university alone, by way of example. Agree that they have a few great schools and millions of talented kids, and I’d wager most would come to Ohio State or whatever (let alone Stanford, Caltech, MIT) than stay at home.
1moretimeagain
Iconic Member
- Messages
- 1,739
China isn’t implementing A-1 learning, so that should give the US a sizable advantage going forward.