FabSwingers.com > Forums > Politics > DeepSeek R1
DeepSeek R1
Jump to: Newest in thread
| |||
| |||
| |||
| |||
"I don’t think anyone knows if it is actually any good yet as it is bowing under the load. It appears to be super cheap to run and that will put a huge hole in the US plan to spend half a trillion dollars on AI." It seems to me that the USA will retain its edge, for now. As you say, it won't be making as much money out of it as it expected to. It's going to be available to poorer countries - with more immediate problems than geopolitics - that might be able to put it to good use. Hopefully, they will. "China bad" is not going to be enough of a reason to prevent others using the opportunity it presents. It's not putting Meta off. | |||
"It might spur change leading to less resource demanding tools, as part of healthy competition. " I think current US policy views the resources required to run current generation AI models as a defensive moat - no one else can afford to play/get access to the compute required. They are going to be _very_ unhappy if someone has come along and blown a hole in that moat. It is pretty good timing from the Chinese as it makes a mockery of the big announcement from the Donald last week where they doubled down on the expenditure. I do wonder if you are going to get big pushback from the US govt and the tech players as they stand to lose a lot if this actually works as claimed. Huge amounts of capital are going to be eviscerated over night. | |||
"It might spur change leading to less resource demanding tools, as part of healthy competition. " Yep. It's out there now. Its very existence, at that price, is going to make competition inevitable. Hopefully, the upside will be bigger than the downside. | |||
"It might spur change leading to less resource demanding tools, as part of healthy competition. I think current US policy views the resources required to run current generation AI models as a defensive moat - no one else can afford to play/get access to the compute required. They are going to be _very_ unhappy if someone has come along and blown a hole in that moat. It is pretty good timing from the Chinese as it makes a mockery of the big announcement from the Donald last week where they doubled down on the expenditure. I do wonder if you are going to get big pushback from the US govt and the tech players as they stand to lose a lot if this actually works as claimed. Huge amounts of capital are going to be eviscerated over night." Agreed. I think pushback is bound to happen - but how? Restricting high end chips doesn't seem to have worked. So, punish the whole world because it doesn't want to buy your stuff? That's Trump's thinking already. It's ironic that the world's economic, free trade, powerhouse is run by a man who thinks tariffs are the answer. I don't suppose we'll have to wait long to see how many US tech firms start using it/modifying it, either, since it's open source. | |||
| |||
"I don't suppose we'll have to wait long to see how many US tech firms start using it/modifying it, either, since it's open source." That is also an amazing thing and quite heartening to hear. That alone will give it wings and democratize the AI revolution for all, and smash the gatekeepers. | |||
"Firstly, I am no expert on AI. I'm just interested in it. After a few hours reading on DeepSeek R1, I'm wondering what impact it might have; beyond upsetting nVidia a bit. Given that it's open source; is it a good thing, or a bad thing?" There are a number of issues, the first one as you point out is Nvidia and other US AI providers, who are investing billions into AI, the idea that a Chinese company can create AI capabilities spending 6 million and with no access to cutting edge chip technology is making the US investments look risky. This also plays out the same way for others such as the UK, only a couple of weeks ago Starmer announced £14 billion of private investment to provide public service AI capabilities. The UK government signing longterm contracts based on current cost assumptions could find us paying many billions more than necessary. The second issue is data security. The DeepSeek app is now the most downloaded app in the US, which, in my opinion, presents a far greater risk than TikTok ever did. People tend to input far more sensitive information into AI interfaces than they would on a social media platform, personal details, business data, contractual information, costs, and even tech security protocols. Allowing this app to dominate unchecked is alarming, and I’m really surprised it has been allowed to run. I think they were caught napping. | |||
"I don't suppose we'll have to wait long to see how many US tech firms start using it/modifying it, either, since it's open source. That is also an amazing thing and quite heartening to hear. That alone will give it wings and democratize the AI revolution for all, and smash the gatekeepers." I think that's the main thing about it. Analogous to owning a car. You'll have to be very rich to own a top end one, but you won't have to be to own one at all. Trump's reaction to it all has been pragmatic. Fair play to him. And it is good for the USA, given their dominance in very high end chips. My only real fear is that, now that obscene profit seems to be off the table, the 1% are going to be looking to huge redundancies, in order to maintain their positions. Anyway, if it was a "Sputnik moment", it might lead to another "Apollo programme". We can but hope. | |||
| |||
"Firstly, I am no expert on AI. I'm just interested in it. After a few hours reading on DeepSeek R1, I'm wondering what impact it might have; beyond upsetting nVidia a bit. Given that it's open source; is it a good thing, or a bad thing? There are a number of issues, the first one as you point out is Nvidia and other US AI providers, who are investing billions into AI, the idea that a Chinese company can create AI capabilities spending 6 million and with no access to cutting edge chip technology is making the US investments look risky. This also plays out the same way for others such as the UK, only a couple of weeks ago Starmer announced £14 billion of private investment to provide public service AI capabilities. The UK government signing longterm contracts based on current cost assumptions could find us paying many billions more than necessary. The second issue is data security. The DeepSeek app is now the most downloaded app in the US, which, in my opinion, presents a far greater risk than TikTok ever did. People tend to input far more sensitive information into AI interfaces than they would on a social media platform, personal details, business data, contractual information, costs, and even tech security protocols. Allowing this app to dominate unchecked is alarming, and I’m really surprised it has been allowed to run. I think they were caught napping. " They were certainly caught napping - and the notion of financial mismanagement of any project in the UK, during this century, is practically a given. As for data security, it's been shown to be iffy AF since Snowden did his thing. Yes, there's still some of it left - where it really matters. I don't think DeepSeek will dominate for long, given US firms will be able to use and adapt it to improve their own AI programs - but it will stop them overcharging anyone. It will be interesting to see how the poorer nations of the world make use of it. "China bad" is a very weak argument where they are concerned. | |||
"Firstly, I am no expert on AI. I'm just interested in it. After a few hours reading on DeepSeek R1, I'm wondering what impact it might have; beyond upsetting nVidia a bit. Given that it's open source; is it a good thing, or a bad thing? There are a number of issues, the first one as you point out is Nvidia and other US AI providers, who are investing billions into AI, the idea that a Chinese company can create AI capabilities spending 6 million and with no access to cutting edge chip technology is making the US investments look risky. This also plays out the same way for others such as the UK, only a couple of weeks ago Starmer announced £14 billion of private investment to provide public service AI capabilities. The UK government signing longterm contracts based on current cost assumptions could find us paying many billions more than necessary. The second issue is data security. The DeepSeek app is now the most downloaded app in the US, which, in my opinion, presents a far greater risk than TikTok ever did. People tend to input far more sensitive information into AI interfaces than they would on a social media platform, personal details, business data, contractual information, costs, and even tech security protocols. Allowing this app to dominate unchecked is alarming, and I’m really surprised it has been allowed to run. I think they were caught napping. They were certainly caught napping - and the notion of financial mismanagement of any project in the UK, during this century, is practically a given. As for data security, it's been shown to be iffy AF since Snowden did his thing. Yes, there's still some of it left - where it really matters. I don't think DeepSeek will dominate for long, given US firms will be able to use and adapt it to improve their own AI programs - but it will stop them overcharging anyone. It will be interesting to see how the poorer nations of the world make use of it. "China bad" is a very weak argument where they are concerned." I think we are seeing a much wider picture emerging. Putting the cost implications to one side, those are just investor wobbles that need calming. AI, as it stands today, is impressive, but it will never reach its full potential without quantum computing. That’s the real game changer, and it would explain why there is so much uneasiness about China’s sudden leap in AI capabilities. Whoever combines those 2 things together first, literally holds the keys to the world. | |||
| |||
| |||
" AI, as it stands today, is impressive, but it will never reach its full potential without quantum computing. That’s the real game changer, and it would explain why there is so much uneasiness about China’s sudden leap in AI capabilities. Whoever combines those 2 things together first, literally holds the keys to the world." Quantum computing and the current generative AI models have nothing practical in common and there is no indication that they are linked in a meaningful way. Large language models are just huge probability based matrices that require standard matrix algebra to run. These matrices are very dense. Quantum computing only really works in very sparse matrices and even then it is subject to huge error rates which means the matrices are very small. Will these issues be solved? Possibly. But not in any meaningful time for the current AI euphoria to deal with and its implications will be in cryptography areas first and that’s huge. | |||
"Firstly, I am no expert on AI. I'm just interested in it. After a few hours reading on DeepSeek R1, I'm wondering what impact it might have; beyond upsetting nVidia a bit. Given that it's open source; is it a good thing, or a bad thing? There are a number of issues, the first one as you point out is Nvidia and other US AI providers, who are investing billions into AI, the idea that a Chinese company can create AI capabilities spending 6 million and with no access to cutting edge chip technology is making the US investments look risky. This also plays out the same way for others such as the UK, only a couple of weeks ago Starmer announced £14 billion of private investment to provide public service AI capabilities. The UK government signing longterm contracts based on current cost assumptions could find us paying many billions more than necessary. The second issue is data security. The DeepSeek app is now the most downloaded app in the US, which, in my opinion, presents a far greater risk than TikTok ever did. People tend to input far more sensitive information into AI interfaces than they would on a social media platform, personal details, business data, contractual information, costs, and even tech security protocols. Allowing this app to dominate unchecked is alarming, and I’m really surprised it has been allowed to run. I think they were caught napping. They were certainly caught napping - and the notion of financial mismanagement of any project in the UK, during this century, is practically a given. As for data security, it's been shown to be iffy AF since Snowden did his thing. Yes, there's still some of it left - where it really matters. I don't think DeepSeek will dominate for long, given US firms will be able to use and adapt it to improve their own AI programs - but it will stop them overcharging anyone. It will be interesting to see how the poorer nations of the world make use of it. "China bad" is a very weak argument where they are concerned. I think we are seeing a much wider picture emerging. Putting the cost implications to one side, those are just investor wobbles that need calming. AI, as it stands today, is impressive, but it will never reach its full potential without quantum computing. That’s the real game changer, and it would explain why there is so much uneasiness about China’s sudden leap in AI capabilities. Whoever combines those 2 things together first, literally holds the keys to the world." Indeed. Let's hope that the United States, should it get there first, stays its hand; as it did when it had the monopoly on nuclear weapons. Possibly a forlorn hope, given the catastrophic decline in the quality of its leadership since 1945. | |||
" AI, as it stands today, is impressive, but it will never reach its full potential without quantum computing. That’s the real game changer, and it would explain why there is so much uneasiness about China’s sudden leap in AI capabilities. Whoever combines those 2 things together first, literally holds the keys to the world. Quantum computing and the current generative AI models have nothing practical in common and there is no indication that they are linked in a meaningful way. Large language models are just huge probability based matrices that require standard matrix algebra to run. These matrices are very dense. Quantum computing only really works in very sparse matrices and even then it is subject to huge error rates which means the matrices are very small. Will these issues be solved? Possibly. But not in any meaningful time for the current AI euphoria to deal with and its implications will be in cryptography areas first and that’s huge." Fair enough. The point I already made about the USA staying its hand, if it gets there first, would still be what I would hope for. As you say, a breakthrough in quantum computing, rendering cryptography useless; that's a very different, even less stable world. | |||
" AI, as it stands today, is impressive, but it will never reach its full potential without quantum computing. That’s the real game changer, and it would explain why there is so much uneasiness about China’s sudden leap in AI capabilities. Whoever combines those 2 things together first, literally holds the keys to the world. Quantum computing and the current generative AI models have nothing practical in common and there is no indication that they are linked in a meaningful way. Large language models are just huge probability based matrices that require standard matrix algebra to run. These matrices are very dense. Quantum computing only really works in very sparse matrices and even then it is subject to huge error rates which means the matrices are very small. Will these issues be solved? Possibly. But not in any meaningful time for the current AI euphoria to deal with and its implications will be in cryptography areas first and that’s huge." I see your point about the technical limitations of quantum computing when it comes to current AI models, that’s what I meant with it hasn’t reached its full potential. My point isn’t that quantum computing and generative AI are directly linked right now but rather that the combination of these two technologies has the potential to redefine global power dynamics. Even if quantum computing’s early applications focus on cryptography, that alone has massive geopolitical implications. Combine that with AI’s ability to process and leverage data at unprecedented scales, and whoever masters both technologies first gains a strategic advantage. It was thought China wouldn’t be an AI player but they are a quantum computing player, that’s changed now. | |||
 | |||
| |||
| |||
"It's a good thing that China has challenged US hegemony in the field of AI. The US thought they had an unassailable technology lead over the rest of the world with AI. In their arrogance, they underestimated China. Now the world has an open source platform to develop and level the playing field." There are loads of high quality open source AI models already - just go to Hugging Face to see them. Facebook open sourced Llama a couple of years ago and Deepseek isn’t necessarily better than that in terms of output. The innovation isn’t the fact it is open source - the innovation is how (relatively) cheaply it could be trained. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data." The same goes for any model - they are all restricted. Trying searching for anything sexual in most models and it will turn you away. You need to take the output of any model with a very large pinch of salt. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data." Not necessarily. The model apparently gives a fair and unbiased answer to questions around Tiananmen Square. In any event, any filters or bias will soon be discovered in an open source model. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data. Not necessarily. The model apparently gives a fair and unbiased answer to questions around Tiananmen Square. In any event, any filters or bias will soon be discovered in an open source model. " It is the information the user provides that is the issue. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data. The same goes for any model - they are all restricted. Trying searching for anything sexual in most models and it will turn you away. You need to take the output of any model with a very large pinch of salt." Exactly, however 18 - 34 year olds show high levels of trust in AI output and tend not to challenge the data. I think I could also make an assumption that the same age group are more likely to download new apps. I would not be surprised if the app get pulled. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data. Not necessarily. The model apparently gives a fair and unbiased answer to questions around Tiananmen Square. In any event, any filters or bias will soon be discovered in an open source model. It is the information the user provides that is the issue." As in data harvesting you mean? I don't think DeepSeek differs from other AI models in that ambition. | |||
"https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan Well. Who knew. Chinese AI attempts to push the CCP party line. Firmly. One reason be very careful of (a) its responses; and (b) what it does with your data. Not necessarily. The model apparently gives a fair and unbiased answer to questions around Tiananmen Square. In any event, any filters or bias will soon be discovered in an open source model. It is the information the user provides that is the issue. As in data harvesting you mean? I don't think DeepSeek differs from other AI models in that ambition." You might be surprised at the data that gets pumped into AI, far more personal and sensitive data than a normal web search. I knew of an exec that would copy and paste whole sections of contracts and then ask AI questions on the contract. I asked him why he was doing that, he said it was far quicker than waiting for legal / others to get back to him, and it also gave him a sense of understanding that he liked. Medical searches with reference to me / I etc etc. | |||
| |||
| |||
"I dont really get the hype, what can deepseek do that a simple google search wont? How dose this assist me in my day-to-day life? " "I dont really get the hype, what can deepseek do that a simple google search wont? How dose this assist me in my day-to-day life? " Typically when users first starting using AI, they will use it as they do the internet for searches, like you mentioned what can it do that Google can't. Some AI will prompt you for another question after your search, or prompt if you want something clarified, this eases the user into a more natural flow of asking for information. At some point you will start to put together a string of questions such as: Tell me about London, what time is best to go, what will the weather be like during that time and is there anything I should avoid during my visit. All basic but the responses back will start to feel natural, this will then lead to further questions and possibly more reliance on AI, all of which is data provided. You might have an email you need to respond to and you will copy and paste or use an integrated tool to ask for a reply, the data is starting to build and build and the reliance on AI begins to build as you have discovered your own personal assistant. Google in comparison is not as dimensional or natural to use. It makes money through advertising, it wants your data too, so it can directly send you the products you might be interested in and charge the advertisers for putting their content under your nose. If DeepSeek has created an AI service that provides the reliance as above, capturing your data at every level for $6 million, they have cracked a big nut in terms of targeting and influencing you directly at a personal level, for peanuts. | |||
"I dont really get the hype, what can deepseek do that a simple google search wont? How dose this assist me in my day-to-day life? " The results are more personalised. I had the the need for a Will for a family member. Used ChatGBD - just explained the requirements in everyday terms and in seconds it churned out a personalised Will covering all special circumstances. Impressive stuff. | |||
"I dont really get the hype, what can deepseek do that a simple google search wont? How dose this assist me in my day-to-day life? The results are more personalised. I had the the need for a Will for a family member. Used ChatGBD - just explained the requirements in everyday terms and in seconds it churned out a personalised Will covering all special circumstances. Impressive stuff." This is what I was talking about  | |||
"It's a good thing that China has challenged US hegemony in the field of AI. The US thought they had an unassailable technology lead over the rest of the world with AI. In their arrogance, they underestimated China. Now the world has an open source platform to develop and level the playing field." It seems to be a bit like generic drugs, that way. Works well enough and doesn't price anyone out of access. Open AI are bleating about it - even though they're as welcome to exploit it as anyone else. What the "outrage" really boils down to is a lost opportunity for the US - the "Broligarchy" in particular - to extort billions of dollars from the rest of the world, in the way they have always felt entitled to. Well, they can suck it up. They'll be fine. | |||
| |||
"Makes you laugh when people say don't trust China with IT, mobiles, social media and now AI, But trust the US or other western governments as they have a proven track record when it comes to not lying snooping on data etc " Yep. It's almost as if Ed Snowden never told anyone anything. | |||
| |||
