sdferr, or anyone else with knowledge of Baltimore. I’m finalizing plans for a trip in mid-April – mostly baseball. If I stay at one of the hotels along the inner harbor, some of them are several blocks from the stadium, is that a safe walk or not a good idea?
I haven’t been in Charm City since 1996-ish, but nevertheless dare to think a few blocks from OPACY ought to be safe enough, Mike. E’en so, further research is warranted.
Maybe query the folks at Roch Kubatco’s daily commentary section here (many of them are residents and go to games frequently). They may even have (salty) recommendations for eateries, hotels, etc. But a warning, it’s Maryland, chock-a-block with leftist doltery.: https://www.masnsports.com/blog/blogger/school-of-roch
Oh, and Nonna rules!
To anyone who is curious, this video has a pretty good and relatively non-technical description of what the new Chinese developed Deepseek R1 is and why it’s significant. But in short, essentially this Chinese company supposedly spent a mere 6 million dollars to develop a very competitive AI by using the far more expensive models to help “train” it. Companies like Meta, Open AI, and Google have spent hundreds of billions to develop their AIs and this little Chinese company used some of those models (the ones that can be downloaded for free on Hugging Face) to develop Deepseek.
“… this little Chinese company used some of those models”
In other words they stole the basics as is so often the case with communist countries. Copyrights mean nothing to them.
In other words they stole the basics
Well, to be completely fair, they didn’t technically do anything illegal by using those models since they’re all open source and free to download for anyone with an internet connection. They just used them in a very clever way, a way which likely represents a paradigm shift in the way these things will be developed in the future.
All that said, since it happened in China, it all feels a little sinister. According to people who have used the version that you can download and run locally on a PC, there doesn’t appear to censorship going on. It’ll responsed to queries about the Tiananmen Square incident for example. I’ve heard that the online version does censor responses, albeit in some strange ways.
Another week more liberal meltdown. This week it’s the executive order calling for a TEMPORARY freeze on federal funding to allow a review. The left is claiming it’s a coup by Trump to start dismantling the federal government to establish a dictatorship.
The insanity continues. I guess we should get use to these psychotic pronouncements.
Deep seek is a Huawei product, remember them, an arm of the PLA, Liang, (lne) escapes me is using NVIDIA chips for the frame,
A few people enjoyed my mention of The Egg and I the other day. The Kettles in the book are as comic, but quite a bit less wholesome, than the Kettles are in the movies.
By the end of the first spring Bob hated the Kettles with a deadly loathing and I couldn’t blame him—they practically doubled his work and certainly impeded his progress. By the time we had weathered the first winter his attitude had softened somewhat, and by the end of the second year he accepted them like one does a birthmark. I enjoyed the Kettles. They shocked, amused, irritated and comforted me. They were never dull and they were always there.
With misfortune constantly stalking them and poverty and confusion always at hand, I was amazed at the harmony that existed among the Kettles. There was no bickering or blaming each other for things that happened—there was no need to, for the fault didn’t lie with them, they figured. Taking great draughts of coffee, Mrs. Kettle told me again and again where the fault lay. “It’s them crooks in Washington,” she said vehemently. “All the time being bribed and buyin’ theirselves big cars with our money.” To Mrs. Kettle there was but one Government and that was in Washington, D. C. She had no knowledge of any county, city or state governments. “The whole damn shebang” was in Washington, and Washington to her was a place where everyone was in full evening dress twenty-four hours a day attending balls and dinners which seethed with spies, crooks, liquor, loose women, Strauss waltzes and bribes. Politics were the Kettles’ out. When the manure in the barn was piled so high Paw couldn’t get in to milk the cows or Tits’ Mervin had given her a black eye, or there was no chicken feed or money to buy any, Mrs. Kettle would say, “Look! Just look what them crooks in Washington has did. They put them new fancy laws on time payments so Paw can’t get a manure spreader. They give Mervin his Indian money so he gits drunk and hits Tits. They’re payin’ the farmers not to raise chicken feed and the price is so high I can’t git the money to buy it. If you want to know what I think,” she would take another strengthening gulp of the coffee, then glaring at Paw, Elwin, Tits and me, would conclude, “I think them politicians can take their crooked laws and their crooked bribes and stuff ’em.” They would all nod wisely. The blame had been put squarely where it belonged and nobody on the Kettle farm had to go sneaking around feeling guilty.
If one can believe AIs – is Deepseek is a Huawei product?
• Grok: No, DeepSeek is not a Huawei product.
• Copilot: No, Deepseek is not a Huawei product.
• ChatGPT: No, DeepSeek is not a Huawei product.
• Claude: DeepSeek is not a Huawei product. DeepSeek is an independent AI research company based in China.
That quoted…it’s still China. I am only testing it on test PCs – Linux was first w/o any GUI success, but terminal was not bad. Just error messages thru browser tests.
DeepSeek has apparently been under attack—especially the registration area and I’ve had no success trying to use it on a browser.
Have prepped ’Rose’ test PC to do some tests w/ Win11 Pro, i.e., I made a backup image to do a full restore if I don’t like the downloads involved in next tests.
“It’s them crooks in Washington.”
And now that we have a group that’s gonna clean up the mess, no more “spies, crooks, liquor, loose women, Strauss waltzes, and bribes” the price of chicken feed will go down. Or something.
Thanks for the morning laugh, NC.
There are claims that DeepSeek’s Terms of Service include mentions of keyboard logging and other sketchy scanning of the computer running it, and sharing those logs with servers in China.
I’ve not gone looking to confirm, and it’s not like there are competitors in that space who might find it useful to spread fear, uncertainty, and doubt. It’s also true that the Chinese (and others) have something of a history of extracting information as payment for otherwise free products.
Yeah, AFAIK the company behind Deepseek-R1 is not owned by Huawei, but it did evidently use Huawei chips to do some of the work as miguel cervantes elucidated. Due to import restrictions, they evidently could only use a slightly older variety of the Nvidia chips to train it and the domestically produced Huawei chips to host it. Those Huawei chips are obviously nowhere near bleeding edge and have extremely low yield rates (around 20% yields compared to around 70% for Western produced chips) making it extremely difficult to produce in large quantities at this time even with all the force of the Chinese government supporting their development. This is due to China’s lack of access to ASMLs most advanced photolithography equipment.
There are claims that DeepSeek’s Terms of Service include mentions of keyboard logging and other sketchy scanning of the computer running it, and sharing those logs with servers in China.
It’s open source, isn’t it? So what’s to stop some enterprising US company from creating a cleaned-up version? Though more likely it would just tweak it to grab information for its own purposes.
CNN anchor Jim Acosta ripped President Donald Trump in a defiant on-air farewell announcing his departure from CNN, telling viewers “This is no time to bow down to a tyrant!”
Some call him China’s Sam Altman … Others compare him to Jim Simons, the pioneer of quantitative investing … Liang Wenfeng shares a lot with both innovators, and his impact might end up being as great.
Artificial-intelligence models from China’s DeepSeek, the company led by Liang, have taken the world by surprise, racing to the global top 10 in both performance and popularity. The company has done it with less-advanced chips than those available in the U.S., jolting technology executives in Silicon Valley, politicians in Washington and investors around the globe.
***
One of Liang’s biggest decisions was to make his code open-source, meaning anyone can access it. He said he wanted DeepSeek to break the monopoly of big tech companies.
***
The sudden surge of people using DeepSeek’s models to pose queries caught Liang and the company off guard, and DeepSeek’s services have repeatedly crashed since Sunday. Liang worked with his team to address the demand before taking a break for the Lunar New Year holiday in China.
It’s open source, isn’t it? So what’s to stop some enterprising US company from creating a cleaned-up version? Though more likely it would just tweak it to grab information for its own purposes.
It’d probably be relatively easy to check to see if the locally run version of it was uploading anything with just packet sniffing, which I have no doubt lots of people have already checked for. Obviously the online version is another matter.
“Wow, really good news!” Trump wrote. “Jim Acosta, one of the worst and most dishonest reporters in journalistic history, a major sleazebag, has been relegated by CNN Fake News to the Midnight hour, ‘Death Valley,’ because of extraordinarily BAD RATINGS (and no talent!).
***
…Brian Stelter then posted Acosta’s statement of resignation.
“I’ve decided to move on” from CNN, @Acosta said, while expressing gratitude to the network and its viewers,” Stelter wrote. “Some parting words: ‘Don’t give in to the lies. Don’t give in to the fear. Hold on to the truth – and to hope.'”
The algorithms the DeepSeek AI uses are open source but the user interface software need not be, and that’s the part collecting your data.
As to the interface, the local (offline) versions of Deepseek will almost always be run on Linux systems and hosted within software called “Ollama”, although there are also versions of it for Windows and Mac. Ollama is an open-source tool developed by an American company. So end-to-end everything is open source. To be clear, that may not completely preclude any and all possible chicanery, but it certainly makes such things far more difficult to do since you have huge numbers of people with a high degree of technical know-how constantly checking things. And of course, this only applies to the locally run model, not the online version.
Looks to me as if they did a very good job of optimization, in addition to having the advantage of previously-created models to build on.
Really, it probably shouldn’t be surprising. Things that gobble up a lot of resources usually get optimized so as to work equally well (or better) while gobbling up less resources.
I’m waiting for the fine print on DeepSeek. The story is too good to be entirely true. The results have been benchmarked, but I am deeply skeptical the Chinese achieved it in two months with a small team for $6 mil using only 2048 Nvidia GPUs.
China is in trouble and they can desperately use some good, better yet spectacular, news on the AI front. Nonetheless, whatever they have done — and they stood on America’s shoulders to do it — we can too. When we figure out their tricks, we will use them.
Ultimately it comes down to compute — the new noun for raw computing power based on processors/GPUs and servers.
We still win.
Thanks, huxley. I can’t claim to understand the technical part at all. But several apparently knowledgeable people on X are pointing out that the Chinese claims about how they did this with so little money and time are not believable and that the functionality is not so wonderful that it should have caused a US market crash.
Well, I can say that 2,048 Nvidia h800’s would cost wayyy more than 6 million USD. Used ones go for like 30k a piece.
Backing Russia fuels Iran, and fueling Iran empowers Russia—two sides of the same coin…
sdferr, or anyone else with knowledge of Baltimore. I’m finalizing plans for a trip in mid-April – mostly baseball. If I stay at one of the hotels along the inner harbor, some of them are several blocks from the stadium, is that a safe walk or not a good idea?
I haven’t been in Charm City since 1996-ish, but nevertheless dare to think a few blocks from OPACY ought to be safe enough, Mike. E’en so, further research is warranted.
Maybe query the folks at Roch Kubatco’s daily commentary section here (many of them are residents and go to games frequently). They may even have (salty) recommendations for eateries, hotels, etc. But a warning, it’s Maryland, chock-a-block with leftist doltery.: https://www.masnsports.com/blog/blogger/school-of-roch
Oh, and Nonna rules!
To anyone who is curious, this video has a pretty good and relatively non-technical description of what the new Chinese developed Deepseek R1 is and why it’s significant. But in short, essentially this Chinese company supposedly spent a mere 6 million dollars to develop a very competitive AI by using the far more expensive models to help “train” it. Companies like Meta, Open AI, and Google have spent hundreds of billions to develop their AIs and this little Chinese company used some of those models (the ones that can be downloaded for free on Hugging Face) to develop Deepseek.
“… this little Chinese company used some of those models”
In other words they stole the basics as is so often the case with communist countries. Copyrights mean nothing to them.
Well, to be completely fair, they didn’t technically do anything illegal by using those models since they’re all open source and free to download for anyone with an internet connection. They just used them in a very clever way, a way which likely represents a paradigm shift in the way these things will be developed in the future.
All that said, since it happened in China, it all feels a little sinister. According to people who have used the version that you can download and run locally on a PC, there doesn’t appear to censorship going on. It’ll responsed to queries about the Tiananmen Square incident for example. I’ve heard that the online version does censor responses, albeit in some strange ways.
Another week more liberal meltdown. This week it’s the executive order calling for a TEMPORARY freeze on federal funding to allow a review. The left is claiming it’s a coup by Trump to start dismantling the federal government to establish a dictatorship.
The insanity continues. I guess we should get use to these psychotic pronouncements.
Deep seek is a Huawei product, remember them, an arm of the PLA, Liang, (lne) escapes me is using NVIDIA chips for the frame,
A few people enjoyed my mention of The Egg and I the other day. The Kettles in the book are as comic, but quite a bit less wholesome, than the Kettles are in the movies.
If one can believe AIs – is Deepseek is a Huawei product?
That quoted…it’s still China. I am only testing it on test PCs – Linux was first w/o any GUI success, but terminal was not bad. Just error messages thru browser tests.
DeepSeek has apparently been under attack—especially the registration area and I’ve had no success trying to use it on a browser.
Have prepped ’Rose’ test PC to do some tests w/ Win11 Pro, i.e., I made a backup image to do a full restore if I don’t like the downloads involved in next tests.
except for that tiny detail
https://x.com/Dorialexander/status/1884201860917256631
“It’s them crooks in Washington.”
And now that we have a group that’s gonna clean up the mess, no more “spies, crooks, liquor, loose women, Strauss waltzes, and bribes” the price of chicken feed will go down. Or something.
Thanks for the morning laugh, NC.
There are claims that DeepSeek’s Terms of Service include mentions of keyboard logging and other sketchy scanning of the computer running it, and sharing those logs with servers in China.
I’ve not gone looking to confirm, and it’s not like there are competitors in that space who might find it useful to spread fear, uncertainty, and doubt. It’s also true that the Chinese (and others) have something of a history of extracting information as payment for otherwise free products.
Yeah, AFAIK the company behind Deepseek-R1 is not owned by Huawei, but it did evidently use Huawei chips to do some of the work as miguel cervantes elucidated. Due to import restrictions, they evidently could only use a slightly older variety of the Nvidia chips to train it and the domestically produced Huawei chips to host it. Those Huawei chips are obviously nowhere near bleeding edge and have extremely low yield rates (around 20% yields compared to around 70% for Western produced chips) making it extremely difficult to produce in large quantities at this time even with all the force of the Chinese government supporting their development. This is due to China’s lack of access to ASMLs most advanced photolithography equipment.
There are claims that DeepSeek’s Terms of Service include mentions of keyboard logging and other sketchy scanning of the computer running it, and sharing those logs with servers in China.
It’s open source, isn’t it? So what’s to stop some enterprising US company from creating a cleaned-up version? Though more likely it would just tweak it to grab information for its own purposes.
1) CNN’s Jim Acosta Rips Trump In Defiant On-Air Announcement He’s Leaving: ‘No Time To Bow Down To A Tyrant’
2) DeepSeek Chief’s Journey From Math Geek to Global Disruptor
It’d probably be relatively easy to check to see if the locally run version of it was uploading anything with just packet sniffing, which I have no doubt lots of people have already checked for. Obviously the online version is another matter.
Also, it seems like they did some very clever/hacky things to really get the most out of the older Nvidia GPUs they had access to. Essentially they bypassed the CUDA software and wrote more custom closer-to-the-metal optimizations. Necessity is often the mother of invention.
The algorithms the DeepSeek AI uses are open source but the user interface software need not be, and that’s the part collecting your data.
More on Jim ‘Major Sleazebag‘ Acosta resigning after President Trump ‘B*#^ch Slapped’ him a few times…
Trump Bids Adieu to ‘Major Sleazebag’ Jim Acosta
As to the interface, the local (offline) versions of Deepseek will almost always be run on Linux systems and hosted within software called “Ollama”, although there are also versions of it for Windows and Mac. Ollama is an open-source tool developed by an American company. So end-to-end everything is open source. To be clear, that may not completely preclude any and all possible chicanery, but it certainly makes such things far more difficult to do since you have huge numbers of people with a high degree of technical know-how constantly checking things. And of course, this only applies to the locally run model, not the online version.
Looks to me as if they did a very good job of optimization, in addition to having the advantage of previously-created models to build on.
Really, it probably shouldn’t be surprising. Things that gobble up a lot of resources usually get optimized so as to work equally well (or better) while gobbling up less resources.
Let’s just listen to some good music:
https://youtu.be/tSt5MaSDDj8?si=eUwqfYl1bUh0RBbC
Re: DeepSeek R1
I’m waiting for the fine print on DeepSeek. The story is too good to be entirely true. The results have been benchmarked, but I am deeply skeptical the Chinese achieved it in two months with a small team for $6 mil using only 2048 Nvidia GPUs.
China is in trouble and they can desperately use some good, better yet spectacular, news on the AI front. Nonetheless, whatever they have done — and they stood on America’s shoulders to do it — we can too. When we figure out their tricks, we will use them.
Ultimately it comes down to compute — the new noun for raw computing power based on processors/GPUs and servers.
We still win.
Thanks, huxley. I can’t claim to understand the technical part at all. But several apparently knowledgeable people on X are pointing out that the Chinese claims about how they did this with so little money and time are not believable and that the functionality is not so wonderful that it should have caused a US market crash.
Well, I can say that 2,048 Nvidia h800’s would cost wayyy more than 6 million USD. Used ones go for like 30k a piece.
Backing Russia fuels Iran, and fueling Iran empowers Russia—two sides of the same coin…
THE RUSSIA-IRAN COALITION DEEPENS
Are those clams?
Dave Begley:
Yes, clams.