Today's weird possible AI bubble pop - brought to you by an AI breakthrough?

In reply to SV reX :

I can't fathom why anyone would download this app to their phone. From a Wired article...

Wired Deepseek

 

What DeepSeek Collects About You

To be clear, DeepSeek is sending your data to China. The English-language DeepSeek privacy policy, which lays out how the company handles user data, is unequivocal: “We store the information we collect in secure servers located in the People's Republic of China.”

In other words, all the conversations and questions you send to DeepSeek, along with the answers that it generates, are being sent to China or can be. DeepSeek’s privacy policies also outline the information it collects about you, which falls into three sweeping categories: information that you share with DeepSeek, information that it automatically collects, and information that it can get from other sources.

I don't understand how people can willingly just accept a Trojan horse. I had a conversation with my teenager about Tic Tok a couple days ago. He didn't understand that harm that it could cause. "I'm just watching videos." He didn't care if China gathered data on him. "What good will data on a random 15 year old do for China?" I explained It's not one teenager, it's almost all of them. And while you don't see any benefit to China in that, they see it that you all won't be just teenagers in the future. You will be in the military, running companies, in politics. You will be the ones in power, and they will have access to all of the stupid things that you have done that they can use against you. And you allowed it to watch stupid videos. 

SV reX said:

There is no such thing as a "privately held startup" in China.

There is, a sufficiently small and uninteresting company that doesn't rely on a lot of government business can avoid attracting the CCP's tentacles. DeepSeek has clearly become far too interesting now though:

https://bigdatachina.csis.org/can-chinese-firms-be-truly-private/

In reply to GameboyRMH :

That's like saying that you have freedom in China as long as nobody sees it. 

In reply to Boost_Crazy :

It must be true if you can post a link. 

In reply to Boost_Crazy :

Not quite. The Chinese government can effectively take control of any company in China and access any of its data, but they don't do this to all companies and the ones they haven't interfered with aren't hiding or doing anything wrong.

In terms of government access to companies' data, sadly the US put a law on the books last year that allows similar access.

Just saw news that (grid) energy company stocks also took a 20~30% hit from this:

https://www.bloomberg.com/news/articles/2025-01-28/deepseek-s-ai-model-just-upended-the-white-hot-us-power-market

DeepSeek is being accused of "Distillation".  Does that mean they just stole OpenAI's intellectual property, and it's not really a break through?

Link to Axios article.

In reply to Indy - Guy :

My first reaction is that it's not the same but remarkably parallel to the complaint that people who actually write words, make images, and compose music have with OpenAI (and most of the existing crop of AI vendors).

"All it's doing is consuming publicly available information about how words are put together or how the world looks."

"Output" is what is created by someone/something that has knowledge/skill. Inferences about how to create new things are drawn from consuming existing output.

I am definitely in the early learning phase on AI, but off the cuff the subject of this complaint sounds almost poetically just. OpenAI's ToS may seek to disallow this, but whether that's valid seems to hinge on the same logic about ownership of creative works that they sometimes seem to trample in training their own models.

In reply to GameboyRMH :

In reply to Boost_Crazy :

Not quite. The Chinese government can effectively take control of any company in China and access any of its data, but they don't do this to all companies and the ones they haven't interfered with aren't hiding or doing anything wrong.

In terms of government access to companies' data, sadly the US put a law on the books last year that allows similar access.

How is that "not quite?" If the Chinese government wants it, it's theirs. Period. 
 

I'm not a fan of the U.S. government collecting my data. But it's not the same thing, from the perspective of a U.S. citizen. It's my own country theoretically looking out for my countries best interests Vs. a foreign economic and potentially military adversary. And if I don't like what the U.S. is doing, I can protest, take it to the courts, or vote for representation that is against those laws. 
 

Try asking about Tiananmen Square in Deepseek and see what you get. 

In reply to SV reX :

From that link...

In the last several years, particularly under Xi Jinping’s rule, there has been an even greater attempt by the political leadership to increase their control over the private sector, reduce its political influence, and ensure its loyalty to the system. This has included expanding the reach of national security policies and regulations. For example, the adoption of the National Intelligence Law in 2017 requires all firms in China to accede to government demands to provide information and data as authorities deem necessary to protect China’s national security. It has also meant using carrots, such as providing industrial policy opportunities to private firms, and sticks, such as the regulatory crackdown on private Internet firms that started in late 2018 and recently concluded. Finally, the CCP has also stepped up efforts to directly influence the corporate governance of private firms, in some cases taking “golden shares” in companies, pushing private firms to form CCP branches (see Figure 3), and integrating firms into the burgeoning “corporate social credit system” (CSCS).

In reply to Boost_Crazy :

"Pay no attention to that man behind the curtain..."

Indy - Guy said:

DeepSeek is being accused of "Distillation".  Does that mean they just stole OpenAI's intellectual property, and it's not really a break through?

Link to Axios article.

They stole OpenAI's data, which OpenAI rightfully stole from multiple other sources

calteg said:

They stole OpenAI's data, which OpenAI rightfully stole from multiple other sources

 

Reminds me of when Apple sued Microsoft for stealing what they had already stolen from Xerox.

calteg said:

Indy - Guy said:

DeepSeek is being accused of "Distillation".  Does that mean they just stole OpenAI's intellectual property, and it's not really a break through?

Link to Axios article.

They stole OpenAI's data, which OpenAI rightfully stole from multiple other sources

Yeah, I get that they stole what was stolen.

Here's what I would like to know: Is DeepSeek truly revolutionary, or is it just a case of lying about what was involved to create it.  Does this stolen "distillation" mean it didn't happen without huge power consumption ( to train the model) and can it be done with lesser chips?

From some of the cursory reading I've done, it sounds like the app is very busily grabbing all sorts of data and sending it back to China:

Luke de Pulford, co-founder of the Inter-Parliamentary Alliance on China (IPAC), shared screenshots from the Chinese AI chatbot’s privacy policy, which stated data it collects is stored in “secure servers located in the People’s Republic of China.” 

“Just fyi, @deepseek_ai collects your IP, keystroke patterns, device info, etc etc, and stores it in China, where all that data is vulnerable to arbitrary requisition from the [Chinese] State,” said de Pulford, leader of IPAC, a global group of lawmakers who seek to hold China accountable for democratic abuses. 

But de Pulford is not the only one warning against it.

“Interested in trying out the DeepSeek AI? I read DeepSeek’s Privacy Policy so you don’t have to,” wrote a popular online attorney. “The DeepSeek privacy policy is: You don’t have any privacy. None. It can collect almost any and all information from you and share it with almost anyone. But don’t worry, all your info is stored safely somewhere in China—where privacy is paramount!”

On top of that, it appears that there's a definite bias to the answers that it gives.  Some of the stuff is obvious, like whistling past the literal graveyards of the Uyghurs and Tiananmen square.  But we've already seen how an unintentional bias can develop in these large language models with the developers working to put guardrails around them to try to prevent it.  If an LLM can be guided in that direction, then why wouldn't it be able to be trained with an intentional bias?  And what bias might the Chinese want to apply to an LLM? 

Given the opportunities that exist with something like this, I can't help but think that there's a proverbial man behind the curtain.  Maybe the power savings aren't there, maybe the training did cost way more and take way longer than they're admitting.  I am not typically a conspiracy theorist, but in this case I'm willing to make a partial exception and say that it's worth considering that there's a huge upside for China if they can upend the AI landscape with technology that they control and they have a unique ability to obfuscate and hide truths.

In reply to wae :

Thanks for sharing that.  Zero chance I'm downloading that App to any of my devices.  ZERO.

VolvoHeretic said:

AI is sure sounding a lot like BitCoin.

I don't think AI is like BitCoin -- rather, AI is like "dot com" in the late 90s. (to pedantic, this is "generative AI", not the more general AI that you think of in science fiction).

Generative AI is software that can do amazing things, stuff that no software before has been able to even approach.  There are undoubtedly a bunch of really useful and valuable things it can be used to accomplish, but the thing is that no one really knows what they are yet.  That's going to take a few years to figure out, and in the mean time there's going to be a lot of investment hype across the board.  Some subset of that investment will turn out to be worth a ton of money and a lot of the rest will be worthless.

This is how "dot com" was in the 90s, contrast Amazon and pets.com for one example.  Both started out as businesses of the "let's sell a specific narrow niche of stuff on the internet" variety (books for Amazon, pet supplies for pets.com), now Amazon is a huge successful company while pets.com is a joke about sock puppets.

BitCoin, by contrast, will never be useful for anything other than speculative bubbles and money laundering.

On the search subtopic - try perplexity.ai.  Provides actual answers to questions as opposed to links to sell you crap.  With reference links so you can sanity check.

I was messing with DeepSeek, Qwen (Alibaba), and ChatGPT last night.  ChatGPT gave the most correct answers to ambiguous questions and followups (Why is there air?).  All three did a pretty damn good job of pre-writing the technical spec I need to get started on.  Qwen was happy to give me router config scripts as a response to a follow up question, which I'm looking forward to trying.  Wish I had this stuff a couple decades ago.

codrus (Forum Supporter) said:

VolvoHeretic said:

AI is sure sounding a lot like BitCoin.

I don't think AI is like BitCoin -- rather, AI is like "dot com" in the late 90s. (to pedantic, this is "generative AI", not the more general AI that you think of in science fiction).

Generative AI is software that can do amazing things, stuff that no software before has been able to even approach.  There are undoubtedly a bunch of really useful and valuable things it can be used to accomplish, but the thing is that no one really knows what they are yet.  That's going to take a few years to figure out, and in the mean time there's going to be a lot of investment hype across the board.  Some subset of that investment will turn out to be worth a ton of money and a lot of the rest will be worthless.

This is how "dot com" was in the 90s, contrast Amazon and pets.com for one example.  Both started out as businesses of the "let's sell a specific narrow niche of stuff on the internet" variety (books for Amazon, pet supplies for pets.com), now Amazon is a huge successful company while pets.com is a joke about sock puppets.

BitCoin, by contrast, will never be useful for anything other than speculative bubbles and money laundering.

 

One thing we've found helpful so far is the ability to take a meeting transcript and have it highlight and make meeting notes. 

When you spend multiple hours per day in meetings, it's incredibly helpful to be able to listen and participate without scribbling down half-assed notes and missing things vs getting a good synopsis in just a few minutes after the meeting is over. 

Ran across a story on a major downside of the new DeepSeek model, it's off-the-charts easy to jailbreak:

https://www.wired.com/story/deepseeks-ai-jailbreak-prompt-injection-attacks/